nom.tam.fits.BinaryTable Maven / Gradle / Ivy
package nom.tam.fits; /*- * #%L * nom.tam FITS library * %% * Copyright (C) 1996 - 2024 nom-tam-fits * %% * This is free and unencumbered software released into the public domain. * * Anyone is free to copy, modify, publish, use, compile, sell, or * distribute this software, either in source code form or as a compiled * binary, for any purpose, commercial or non-commercial, and by any * means. * * In jurisdictions that recognize copyright laws, the author or authors * of this software dedicate any and all copyright interest in the * software to the public domain. We make this dedication for the benefit * of the public at large and to the detriment of our heirs and * successors. We intend this dedication to be an overt act of * relinquishment in perpetuity of all present and future rights to this * software under copyright law. * * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. * IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR ANY CLAIM, DAMAGES OR * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR * OTHER DEALINGS IN THE SOFTWARE. * #L% */ import java.io.IOException; import java.lang.reflect.Array; import java.math.BigDecimal; import java.math.BigInteger; import java.text.DecimalFormat; import java.text.ParsePosition; import java.util.ArrayList; import java.util.Arrays; import java.util.List; import java.util.StringTokenizer; import java.util.logging.Logger; import nom.tam.fits.header.Bitpix; import nom.tam.fits.header.NonStandard; import nom.tam.fits.header.Standard; import nom.tam.util.ArrayDataInput; import nom.tam.util.ArrayDataOutput; import nom.tam.util.ArrayFuncs; import nom.tam.util.AsciiFuncs; import nom.tam.util.ColumnTable; import nom.tam.util.ComplexValue; import nom.tam.util.Cursor; import nom.tam.util.FitsEncoder; import nom.tam.util.FitsIO; import nom.tam.util.Quantizer; import nom.tam.util.RandomAccess; import nom.tam.util.ReadWriteAccess; import nom.tam.util.TableException; import nom.tam.util.type.ElementType; import edu.umd.cs.findbugs.annotations.SuppressFBWarnings; /** * Table data for binary table HDUs. It has been thoroughly re-written for 1.18 to improve consistency, increase * performance, make it easier to use, and to enhance. * * @see BinaryTableHDU * @see AsciiTable */ @SuppressWarnings("deprecation") public class BinaryTable extends AbstractTableData implements Cloneable { /** For fixed-length columns */ private static final char POINTER_NONE = 0; /** FITS 32-bit pointer type for variable-sized columns */ private static final char POINTER_INT = 'P'; /** FITS 64-bit pointer type for variable-sized columns */ private static final char POINTER_LONG = 'Q'; /** Shape firs singleton / scalar entries */ private static final int[] SINGLETON_SHAPE = new int[0]; /** The substring convention marker */ private static final String SUBSTRING_MARKER = ":SSTR"; /** * Describes the data type and shape stored in a binary table column. */ public static class ColumnDesc implements Cloneable { private boolean warnedFlatten; /** byte offset of element from row start */ private int offset; /** The number of primitive elements in the column */ private int fitsCount; /** The dimensions of the column */ private int[] fitsShape = SINGLETON_SHAPE; /** Shape on the Java side. Differs from the FITS TDIM shape for String and complex values. */ private int[] legacyShape = SINGLETON_SHAPE; /** Length of string elements */ private int stringLength = -1; /** The class array entries on the Java side. */ private Class> base; /** The FITS element class associated with the column. */ private Class> fitsBase; /** Heap pointer type actually used for locating variable-length column data on the heap */ private char pointerType; /** * String component delimiter for substring arrays, for example as defined by the TFORM keyword that uses the * substring array convention... */ private byte delimiter; /** * Is this a complex column. Each entry will be associated with a float[2] or double[2] */ private boolean isComplex; /** * Whether this column contains bit arrays. These take up to 8-times less space than logicals, which occupy a * byte per value. */ private boolean isBits; /** * User defined column name */ private String name; private Quantizer quant; /** * Creates a new column descriptor with default settings and 32-bit integer heap pointers. */ protected ColumnDesc() { } /** * Creates a new column descriptor with default settings, and the specified type of heap pointers * * @param type The Java type of base elements that this column is designated to contain. For example *
type columns will convert the string provided the string's length * does not exceed the entry size for these columns (the array elements will be padded with zeroes). Note, that * scalarint.class
if the column will contain integers or arrays of integers. * * @throws FitsException if the base type is not one that can be used in binary table columns. */ private ColumnDesc(Class> type) throws FitsException { this(); base = type; if (base == boolean.class) { fitsBase = byte.class; isBits = true; } else if (base == Boolean.class) { base = boolean.class; fitsBase = byte.class; } else if (base == String.class) { fitsBase = byte.class; } else if (base == ComplexValue.class) { base = double.class; fitsBase = double.class; isComplex = true; } else if (base == ComplexValue.Float.class) { base = float.class; fitsBase = float.class; isComplex = true; } else if (base.isPrimitive()) { fitsBase = type; if (base == char.class && FitsFactory.isUseUnicodeChars()) { LOG.warning("char[] will be written as 16-bit integers (type 'I'), not as a ASCII bytes (type 'A')" + " in the binary table. If that is not what you want, you should set FitsFactory.setUseUnicodeChars(false)."); LOG.warning( "Future releases will disable Unicode support by default as it is not supported by the FITS standard." + " If you do want it still, use FitsFactory.setUseUnicodeChars(true) explicitly to keep the non-standard " + " behavior as is."); } } else { throw new TableException("Columns of type " + base + " are not supported."); } } /** * Creates a new column descriptor for the specified boxed Java type, and fixed array shape. The type may be any * primitive type, or elseString.class
,Boolean.class
(for FITS logicals), *ComplexValue.class
orComplexValue.Float.class
(for complex values with 64-bit and * 32-bit precision, respectively). Whereas {@link Boolean} type columns will be stored as FITS logicals (1 * element per byte),boolean
types will be stored as packed bits (with up to 8 bits per byte). * * @param base The Java type of base elements that this column is designated to contain. For example *int.class
if the column will contain integers or arrays of integers. * @param dim the fixed dimensions of the table entries. For strings the trailing dimension must * specify the fixed length of strings. * * @throws FitsException if the base type is not one that can be used in binary table columns. * * @see #createForScalars(Class) * @see #createForStrings(int) * @see #createForStrings(int, int[]) * @see #createForVariableSize(Class) * * @since 1.18 */ public ColumnDesc(Class> base, int... dim) throws FitsException { this(base); setBoxedShape(dim); } /** * Sets a user-specified name for this column. The specified name will be used as the TTYPEn value for this * column. * * @param value The new name for this column. * * @return itself, to support builder patterns. * * @throws IllegalArgumentException If the name contains characters outside of the ASCII range of 0x20 - 0x7F * allowed by FITS. * * @see #name() * @see #getDescriptor(String) * @see #indexOf(String) * @see #addColumn(ColumnDesc) * * @since 1.20 * * @author Attila Kovacs */ public ColumnDesc name(String value) throws IllegalArgumentException { HeaderCard.validateChars(value); this.name = value; return this; } /** * Returns the name of this column, as it was stored or would be stored by a TTYPEn value in the FITS header. * * @return the name of this column * * @see #name(String) * * @since 1.20 * * @author Attila Kovacs */ public String name() { return this.name; } /** * Returns the conversion between decimal and integer data representations for the column data. * * @return the quantizer that converts between floating-point and integer data representations, which may be *null
. * * @see #setQuantizer(Quantizer) * * @since 1.20 */ public Quantizer getQuantizer() { return quant; } /** * Sets the conversion between decimal and integer data representations for the column data. If the table is * read from a FITS input, the column's quantizer is automatically set if the Table HDU's header defines any of * the TSCALn, TZEROn, or TNULLn keywords for the column. Users can override that by specifying another quatizer * to use for the column, or dicard qunatizing by calling this method with anull
argument. * * @param q the quantizer that converts between floating-point and integer data representations, or *null
to not use any quantization, and instead rely on the generic rounding for * decimal-integer conversions for this column. * * @see #getQuantizer() * * @since 1.20 */ public void setQuantizer(Quantizer q) { this.quant = q; } /** * Recalculate the FITS element count based on the shape of the data */ private void calcFitsCount() { fitsCount = 1; for (int size : fitsShape) { fitsCount *= size; } } /** * Sets the shape of entries in the older Java array format of this library (used exclusively prior to 1.18). * For complex columns, there is an extra[2]
dimension, such that single complex values are stored * as an array of[2]
, and an array of n complex values are stored as arrays of *[n][2]
. Otherwise it's the same as {@link #setBoxedShape(int...)}. * * @param dim The Java dimensions for legacy arrays, such as returned by * {@link BinaryTable#getElement(int, int)} */ private void setLegacyShape(int... dim) { legacyShape = dim; calcFitsShape(); calcFitsCount(); } /** * Sets the shape of entries as stored in the FITS header by hte TDIM keyword. * * @param dim The dimensions for the TDIM keyword in Java order (outer dimensions first), which is the reverse * of FITS order (inner-dimensions first). */ private void setFitsShape(int... dim) { fitsShape = dim; calcLegacyShape(); calcFitsCount(); } /** * Sets the shape of boxed Java array entries. For complex columns, all single entries, including strings and * complex values, have scalar shape[]
, whereas an array of n have shape[n]
. * * @param dim The Java dimensions for legacy arrays, such as returned by * {@link BinaryTable#getElement(int, int)} */ private void setBoxedShape(int... dim) { if (isComplex()) { setFitsShape(dim); } else { setLegacyShape(dim); } } /** * Returns the maximum length of string elements contained in this colum, or -1 if this is not a string based * column, or if there is no limit set to string size (e.g. in a variable-length column) * * @return the maximum length of string values stored in this column, or -1 if it is not as string column, or if * its a string column containing strings of unconstrained variable length. * * @see #getEntryShape() * * @since 1.18 */ public final int getStringLength() { return stringLength; } /** * Sets the maximum length of string elements in this column. * * @param len The fixed string length in bytes. */ private void setStringLength(int len) { stringLength = len; if (!isVariableSize()) { calcFitsShape(); calcFitsCount(); } } /** * Returns the string delimiter that separates packed substrings in variable-length string arrays. * * @return the delimiter byte value (usually between 0x20 and 0x7e) or 0 if no delimiter was set. * * @see #getStringLength() */ public final byte getStringDelimiter() { return delimiter; } /** * Creates a new column descriptor for a non-string based scalar column. The type may be any primitive type, or *Boolean.class
(for FITS logicals),ComplexValue.class
or *ComplexValue.Float.class
(for complex values with 64-bit and 32-bit precision, respectively). * Whereas {@link Boolean} type columns will be stored as FITS logicals (1 element per byte), *boolean
types will be stored as packed bits (with up to 8 bits per byte). * * @param type The Java type of base elements that this column is designated to contain. * For exampleint.class
if the column will contain integers * or arrays of integers. It must not beString.class
. To * create scalar {@link String} columns use {@link #createForStrings(int)} * instead. * * @return the new column descriptor. * * @throws IllegalArgumentException if the type isString.class
, for which you should be using * {@link #createForStrings(int)} instead. * @throws FitsException if the base type is not one that can be used in binary table columns. * * @see #createForFixedArrays(Class, int[]) * @see #createForVariableSize(Class) * * @since 1.18 */ public static ColumnDesc createForScalars(Class> type) throws IllegalArgumentException, FitsException { if (String.class.isAssignableFrom(type)) { throw new IllegalArgumentException("Use the createStrings(int) method for scalar strings."); } return new ColumnDesc(type, SINGLETON_SHAPE); } /** * Creates a new column descriptor for fixed-shape non-string arrays. The type may be any primitive type, or * elseBoolean.class
(for FITS logicals),ComplexValue.class
or *ComplexValue.Float.class
(for complex values with 64-bit and 32-bit precision, respectively). * Whereas {@link Boolean} type columns will be stored as FITS logicals (1 element per byte), *boolean
types will be stored as packed bits (with up to 8 bits per byte). * * @param type The Java type of base elements that this column is designated to contain. * For exampleint.class
if the column will contain integers * or arrays of integers. It must not beString.class
. To * create scalar {@link String} columns use {@link #createForStrings(int)} * instead. * @param dim the fixed dimensions of the table entries. For strings the trailing * dimension must specify the fixed length of strings. * * @return the new column descriptor. * * @throws IllegalArgumentException if the type isString.class
, for which you should be using * {@link #createForStrings(int, int[])} instead. * @throws FitsException if the base type is not one that can be used in binary table columns. * * @see #createForScalars(Class) * @see #createForStrings(int) * @see #createForStrings(int, int[]) * @see #createForVariableSize(Class) * * @since 1.18 */ public static ColumnDesc createForFixedArrays(Class> type, int... dim) throws IllegalArgumentException, FitsException { if (String.class.isAssignableFrom(type)) { throw new IllegalArgumentException("Use the createStrings(int) method for scalar strings."); } return new ColumnDesc(type, dim); } /** * Creates a new column descriptor for single string entries of fixed maximum length. * * @param len The fixed string length in bytes. * * @return the new column descriptor * * @throws FitsException if the base type is not one that can be used in binary table columns. * * @see #createForScalars(Class) * @see #createForStrings(int, int[]) * * @since 1.18 */ public static ColumnDesc createForStrings(int len) throws FitsException { return createForStrings(len, SINGLETON_SHAPE); } /** * Creates a new column descriptor for arrays of string entries of fixed maximum length. * * @param len The fixed string length in bytes. * @param outerDims The shape of string arrays * * @return the new column descriptor * * @throws FitsException if the base type is not one that can be used in binary table columns. * * @see #createForVariableStringArrays(int) * @see #createForStrings(int) * * @since 1.18 */ public static ColumnDesc createForStrings(int len, int... outerDims) throws FitsException { ColumnDesc c = new ColumnDesc(String.class); c.setLegacyShape(outerDims); c.setStringLength(len); return c; } /** * Creates a new column descriptor for variable-length arrays of fixed-length string entries. Each string * component will occupy exactlylen
bytes. * * @param len The fixed string storage length in bytes. * * @return the new column descriptor * * @throws FitsException if the column could not be created. * * @see #createForDelimitedStringArrays(byte) * @see #createForStrings(int, int[]) * * @since 1.18 */ public static ColumnDesc createForVariableStringArrays(int len) throws FitsException { ColumnDesc c = createForVariableSize(String.class); c.setStringLength(len); return c; } /** * Creates a new column descriptor for variable-length arrays of delimited string entries. * * @param delim the byte value that delimits strings that are shorter than the storage length. It * should be in the ASCII range of 0x20 through 0x7e. * * @return the new column descriptor * * @throws FitsException if the column could not be created. * * @see #createForDelimitedStringArrays(byte) * @see #createForStrings(int, int[]) * * @since 1.18 */ public static ColumnDesc createForDelimitedStringArrays(byte delim) throws FitsException { ColumnDesc c = createForVariableStringArrays(-1); c.setStringDelimiter(delim); return c; } /** * Creates a new column descriptor for variable length 1D arrays or strings. The type may be any primitive type, * or elseString.class
,Boolean.class
(for FITS logicals), *ComplexValue.class
orComplexValue.Float.class
(for complex values with 64-bit and * 32-bit precision, respectively). Whereas {@link Boolean} type columns will be stored as FITS logicals (1 * element per byte),boolean
types will be stored as packed bits (with up to 8 elements per byte). * * @param type The Java type of base elements that this column is designated to contain. For example *int.class
if the column will contain integers or arrays of integers. * * @return the new column descriptor * * @throws FitsException if the base type is not one that can be used in binary table columns. * * @see #createForScalars(Class) * @see #createForStrings(int) * @see #createForStrings(int, int[]) * @see #ColumnDesc(Class, int[]) * * @since 1.18 */ public static ColumnDesc createForVariableSize(Class> type) throws FitsException { ColumnDesc c = new ColumnDesc(type); c.setVariableSize(false); return c; } /** * Recalculate the legacy Java entry shape from the FITS shape (as stored by TDIM). Strings drop the last * dimension from the FITS shape (which becomes the string length), while complex values add a dimension of *[2]
to the FITS shape, reflecting the shape of their real-valued components. */ private void calcLegacyShape() { if (isString()) { legacyShape = Arrays.copyOf(fitsShape, fitsShape.length - 1); stringLength = fitsShape[fitsShape.length - 1]; } else if (isComplex()) { legacyShape = Arrays.copyOf(fitsShape, fitsShape.length + 1); legacyShape[fitsShape.length] = 2; } else { legacyShape = fitsShape; } } /** * Recalculate the FITS storage shape (as reported by TDIM) from the legacy Java array shape */ private void calcFitsShape() { if (isString()) { fitsShape = Arrays.copyOf(legacyShape, legacyShape.length + 1); fitsShape[legacyShape.length] = stringLength; } else if (isComplex()) { fitsShape = Arrays.copyOf(legacyShape, legacyShape.length - 1); } else { fitsShape = legacyShape; } } /** * Returns the size of table entries in their trailing dimension. * * @return the number of elemental components in the trailing dimension of table entries. * * @see #getLeadingShape() */ private int getLastFitsDim() { return fitsShape[fitsShape.length - 1]; } @Override public ColumnDesc clone() { try { ColumnDesc copy = (ColumnDesc) super.clone(); fitsShape = fitsShape.clone(); legacyShape = legacyShape.clone(); // Model should not be changed... return copy; } catch (CloneNotSupportedException e) { return null; } } /** * Specifies that this columns contains single (not array) boxed entrie, such as single primitives, strings, or * complex values. */ private void setSingleton() { setBoxedShape(SINGLETON_SHAPE); } /** * Checks if this column contains single (scalar / non-array) elements only, including single strings or single * complex values. * * @returntrue
if the column contains individual elements of its type, or elsefalse
* if it contains arrays. * * @since 1.18 */ public final boolean isSingleton() { if (isVariableSize()) { return isString() ? (stringLength < 0 && delimiter == 0) : false; } if (isComplex()) { return fitsShape.length == 0; } return legacyShape.length == 0; } /** * Checks if this column contains logical values. FITS logicals can each hvetrue
, *false
ornull
(undefined) values. It is the support for these undefined values that * set it apart from typical booleans. Also, logicals are stored as one byte per element. So if using only *true
,false
values withoutnull
bits will offer more compact storage * (by up to a factor of 8). You can convert existing logical columns to bits via * {@link BinaryTable#convertToBits(int)}. * * @returntrue
if this column contains logical values. * * @see #isBits() * @see BinaryTable#convertToBits(int) * * @since 1.18 */ public final boolean isLogical() { return base == Boolean.class || (base == boolean.class && !isBits); } /** * Checks if this column contains only true boolean values (bits). Unlike logicals, bits can have only *true
,false
values with no support fornull
, but offer more compact * storage (by up to a factor of 8) than logicals. You can convert existing logical columns to bits via * {@link BinaryTable#convertToBits(int)}. * * @returntrue
if this column containstrue
/false
bits only. * * @see #isLogical() * @see BinaryTable#convertToBits(int) * * @since 1.18 */ public final boolean isBits() { return base == boolean.class && isBits; } /** * Checks if this column stores ASCII strings. * * @returntrue
if this column contains only strings. * * @see #isVariableSize() */ public final boolean isString() { return base == String.class; } /** * Checks if this column contains complex values. You can convert suitable columns offloat
or *double
elements to complex using {@link BinaryTable#setComplexColumn(int)}, as long as the last * dimension is 2, ir if the variable-length columns contain even-number of values exclusively. * * @returntrue
if this column contains complex values. */ public final boolean isComplex() { return isComplex; } /** * Checks if this column contains numerical values, such as any primitive number type (e.g. *nt.class
ordouble.class
) or else a {@link ComplexValue} type. type. * * @returntrue
if this column contains numerical data, including complex-valued data. String, * bits, and FITS logicals are not numerical (but all other column types are). * * @since 1.20 */ public final boolean isNumeric() { return !isLogical() && !isBits() && !isString(); } /** * Returns the Java array element type that is used in Java to represent data in this column. When accessing * columns or their elements in the old way, through arrays, this is the type that arrays from the Java side * will expect or provide. For example, when storing {@link String} values (regular or variable-sized), this * will returnString.class
. Arrays returned by {@link BinaryTable#getColumn(int)}, * {@link BinaryTable#getRow(int)}, and {@link BinaryTable#getElement(int, int)} will return arrays of this * type, and the equivalent methods for setting data will expect arrays of this type as their argument. * * @return the Java class, arrays of which, packaged data for this column on the Java side. * * @deprecated Ambiguous, use {@link #getLegacyBase()} instead. It can be confusing since it is not clear if it * refers to array element types used in FITS storage or on the java side when using the older * array access, or if it refers to the class of entries in the main table, which may be heap * pointers. It is also distinct from {@link #getElementClass()}, which returns the boxed type * used by {@link BinaryTable#get(int, int)} or {@link BinaryTable#set(int, int, Object)}. */ public Class> getBase() { return getLegacyBase(); } /** * Returns the primitive type that is used to store the data for this column in the FITS representation. This is * the class for the actual data type, whether regularly shaped (multidimensional) arrays or variable length * arrays (on the heap). For example, when storing {@link String} values (regular or variable-sized), this will * returnbyte.class
. * * @return the primitive class, in used for storing data in the FITS representation. * * @see #getLegacyBase() * * @since 1.18 */ final Class> getFitsBase() { return fitsBase; } /** ** Returns the Java array element type that is used in Java to represent data in this column for the legacy * table access methods. When accessing columns or their elements in the old way, through arrays, this is the * type that arrays from the Java side will expect or provide. For example, when storing complex values (regular * or variable-sized), this will return
*float.class
ordouble.class
. Arrays returned * by {@link BinaryTable#getColumn(int)}, {@link BinaryTable#getRow(int)}, and * {@link BinaryTable#getElement(int, int)} will return arrays of this type. ** This is different from {@link #getElementClass()}, which in turn returns the boxed type of objects returned * by {@link BinaryTable#get(int, int)}. * * @return the Java class, arrays of which, packaged data for this column on the Java side. * * @see #getElementClass() * * @since 1.18 */ public Class> getLegacyBase() { return base; } /** * (for internal use) Returns the primitive data class which is used for storing entries in the main * (regular) table. For variable-sized columns, this will be the heap pointer class, not the FITS data class. * * @return the class in which main table entries are stored. * * @see #isVariableSize() */ private Class> getTableBase() { return isVariableSize() ? pointerClass() : getFitsBase(); } /** * Returns the dimensions of elements in this column. As of 1.18, this method returns a copy ot the array used * internally, which is safe to modify. * * @return an array with the element dimensions. * * @deprecated (for internal use) Use {@link #getEntryShape()} instead. Not useful to users since it * returns the dimensions of the primitive storage types, which is not always the dimension of * table entries on the Java side. */ public int[] getDimens() { return fitsShape.clone(); } /** * (for internal use) The dimension of elements in the FITS representation. * * @return the dimension of elements in the FITS representation. For example an array of string will be 2 * (number of string, number of bytes per string). * * @see #getEntryDimension() */ private int fitsDimension() { return fitsShape.length; } /** * Returns the boxed Java type of elements stored in a column. * * @return The java type of elements in the columns. For columns containing strings, FITS logicals, or complex * values it will be
String.class
,Boolean.class
or *ComplexValue.class
respectively. For all other column types the primitive class of * the elements contained (e.g.char.class
,float.class
) is returned. * * @since 1.18 * * @see ColumnDesc#getElementCount() * @see ColumnDesc#getEntryShape() * @see ColumnDesc#getLegacyBase() */ public final Class> getElementClass() { if (isLogical()) { return Boolean.class; } if (isComplex()) { return ComplexValue.class; } return base; } /** * Returns the dimensionality of the 'boxed' elements as returned by {@link BinaryTable#get(int, int)} or * expected by {@link BinaryTable#set(int, int, Object)}. That is it returns the dimnesion of 'boxed' elements, * such as strings or complex values, rather than the dimension of characters or real components stored in the * FITS for these. * * @return the number of array dimensions in the 'boxed' Java type for this column. Variable-sized columns will * always return 1. * * @see #getEntryShape() * @see #getElementCount() * * @since 1.18 */ public final int getEntryDimension() { if (isVariableSize()) { return 1; } return isString() ? legacyShape.length : fitsShape.length; } /** * Returns the array shape of the 'boxed' elements as returned by {@link BinaryTable#get(int, int)} or expected * by {@link BinaryTable#set(int, int, Object)}. That is it returns the array shape of 'boxed' elements, such as * strings or complex values, rather than the shape of characters or real components stored in the FITS for * these. * * @return the array sized along each of the dimensions in the 'boxed' Java type for this column, or *null
if the data is stored as variable-sized one-dimensional arrays of the boxed * element type. (Note, that accordingly variable-length string columns containing single strings * will thus return{1}
, notnull
). * * @see #getEntryShape() * @see #getElementCount() * @see #isVariableSize() * * @since 1.18 */ public final int[] getEntryShape() { if (isVariableSize()) { return null; } if (isComplex) { return fitsShape.clone(); } return legacyShape.clone(); } /** * Returns the number of primitive elements (sych as bytes) that constitute a Java element (such as a String) in * this table. * * @return The number of primitives per Java element in the column, that is 1 for columns of primitive types, 2 * for complex-valued columns, or the number of bytes (characters) in a String element. * Variable-length strings will return -1. * * @since 1.18 * * @see #getElementCount() * @see #getLegacyBase() */ public final int getElementWidth() { if (isComplex()) { return 2; } if (isString()) { return getStringLength(); } return 1; } /** * Returns the number of 'boxed' elements as returned by {@link BinaryTable#get(int, int)} or expected by * {@link BinaryTable#set(int, int, Object)}. That is it returns the number of strings or complex values per * table entry, rather than the number of of characters or real components stored in the FITS for these. * * @return the number of array elements in the 'boxed' Java type for this column, or -1 if the column contains * elements of varying size. * * @see #getEntryShape() * @see #getEntryDimension() * @see #isVariableSize() * * @since 1.18 */ public final int getElementCount() { if (isVariableSize()) { return isString() ? 1 : -1; } if (isString()) { return fitsCount / getStringLength(); } return fitsCount; } /** * Returns the number of primitive base elements for a given FITS element count. * * @param fitsLen the FITS element count, sucj a a number of integers, complex-values, or bits * * @return the number of Java primitives that will be used to represent the number of FITS values for * this type of column. * * @see #getFitsBase() */ private int getFitsBaseCount(int fitsLen) { if (isBits) { return (fitsLen + Byte.SIZE - 1) / Byte.SIZE; } if (isComplex) { return fitsLen << 1; } return fitsLen; } /** * Returns the number of regular primitive table elements in this column. For example, variable-length columns * will always return 2, and complex-valued columns will return twice the number of complex values stored in * each table entry. * * @return the number of primitive table elements * * @since 1.18 */ public final int getTableBaseCount() { if (isVariableSize()) { return 2; } return getFitsBaseCount(fitsCount); } /** * Checks if this column contains entries of different size. Data for variable length coulmns is stored on the * heap as one-dimemnsional arrays. As such information about the 'shape' of data is lost when they are stored * that way. * * @returntrue
if the column contains elements of variable size, or elsefalse
if all * entries have the same size and shape. */ public final boolean isVariableSize() { return pointerType != POINTER_NONE; } /** * @deprecated (for internal use) This method should be private in the future. * * @return new instance of the array with space for the specified number of rows. * * @param nRow the number of rows to allocate the array for */ public Object newInstance(int nRow) { return ArrayFuncs.newInstance(getTableBase(), getTableBaseCount() * nRow); } /** * @deprecated (for internal use) It may be reduced to private visibility in the future. Returns the * number of bytes that each element occupies in its FITS serialized form in the stored row * data. * * @return the number of bytes an element occupies in the FITS binary table data representation */ public int rowLen() { return getTableBaseCount() * ElementType.forClass(getTableBase()).size(); } /** * Checks if this column used 64-bit heap pointers. * * @returntrue
if the column uses 64-bit heap pointers, otherwisefalse
* * @see #createForVariableSize(Class) * * @since 1.18 */ public boolean hasLongPointers() { return pointerType == POINTER_LONG; } /** * Returns theTFORM
n character code for the heap pointers in this column or 0 if this is * not a variable-sized column. * * @returnint.class
orlong.class
*/ private char pointerType() { return pointerType; } /** * Returns the primitive class used for sotring heap pointers for this column * * @returnint.class
orlong.class
*/ private Class> pointerClass() { return pointerType == POINTER_LONG ? long.class : int.class; } /** * Sets whether this column will contain variable-length data, rather than fixed-shape data. * * @param useLongPointerstrue
to use 64-bit heap pointers for variable-length arrays or else *false
to use 32-bit pointers. */ private void setVariableSize(boolean useLongPointers) { pointerType = useLongPointers ? POINTER_LONG : POINTER_INT; fitsCount = 2; fitsShape = new int[] {2}; legacyShape = fitsShape; stringLength = -1; } /** * Sets a custom substring delimiter byte for variable length string arrays, between ASCII 0x20 and 0x7e. We * will however tolerate values outside of that range, but log an appropriate warning to alert users of the * violation of the standard. User's can either 'fix' it, or suppress the warning if they want to stick to their * guns. * * @param delim the delimiter byte value, between ASCII 0x20 and 0x7e (inclusive). * * @since 1.18 */ private void setStringDelimiter(byte delim) { if (delim < FitsUtil.MIN_ASCII_VALUE || delim > FitsUtil.MAX_ASCII_VALUE) { LOG.warning("WARNING! Substring terminator byte " + (delim & FitsIO.BYTE_MASK) + " outside of the conventional range of " + FitsUtil.MIN_ASCII_VALUE + " through " + FitsUtil.MAX_ASCII_VALUE + " (inclusive)"); } delimiter = delim; } /** * Checks ifnull
array elements are permissible for this column. It is for strings (which map to * empty strings), and for logical columns, where they signify undefined values. * * @returntrue
ifnull
entries are considered valid for this column. */ private boolean isNullAllowed() { return isLogical() || isString(); } /** * Parses the substring array convention from a TFORM value, to set string length (if desired) and a delimiter * in variable-length string arrays. * * @param tform the TFORM header value for this column * @param pos the parse position immediately after the 'A' * @param setLength Whether to use the substring definition to specify the max string component length, for * example because it is not defined otherwise by TDIM. */ private void parseSubstringConvention(String tform, ParsePosition pos, boolean setLength) { if (setLength) { // Default string length... setStringLength(isVariableSize() ? -1 : fitsCount); } // Parse substring array convention... if (pos.getIndex() >= tform.length()) { return; } // Try 'rAw' format... try { int len = AsciiFuncs.parseInteger(tform, pos); if (setLength) { setStringLength(len); } return; } catch (Exception e) { // Keep going... } // Find if and where is the ":SSTR" marker in the format int iSub = tform.indexOf(SUBSTRING_MARKER, pos.getIndex()); if (iSub < 0) { // No substring definition... return; } pos.setIndex(iSub + SUBSTRING_MARKER.length()); // Set the substring width.... try { int len = AsciiFuncs.parseInteger(tform, pos); if (setLength) { setStringLength(len); } } catch (Exception e) { LOG.warning("WARNING! Could not parse substring length from TFORM: [" + tform + "]"); } // Parse substring array convention... if (pos.getIndex() >= tform.length()) { return; } if (AsciiFuncs.extractChar(tform, pos) != '/') { return; } try { setStringDelimiter((byte) AsciiFuncs.parseInteger(tform, pos)); } catch (NumberFormatException e) { // Warn if the delimiter is outside of the range supported by the convention. LOG.warning("WARNING! Could not parse substring terminator from TFORM: [" + tform + "]"); } } private void appendSubstringConvention(StringBuffer tform) { if (getStringLength() > 0) { tform.append(SUBSTRING_MARKER); tform.append(getStringLength()); if (delimiter != 0) { tform.append('/'); tform.append(new DecimalFormat("000").format(delimiter & FitsIO.BYTE_MASK)); } } } /** * Returns the TFORM header value to use for this column. * * @return The TFORM value that describes this column * * @throws FitsException If the column itself is invalid. */ String getTFORM() throws FitsException { StringBuffer tform = new StringBuffer(); tform.append(isVariableSize() ? "1" + pointerType() : fitsCount); if (base == int.class) { tform.append('J'); } else if (base == short.class) { tform.append('I'); } else if (base == byte.class) { tform.append('B'); } else if (base == char.class) { if (FitsFactory.isUseUnicodeChars()) { tform.append('I'); } else { tform.append('A'); } } else if (base == float.class) { tform.append(isComplex() ? 'C' : 'E'); } else if (base == double.class) { tform.append(isComplex() ? 'M' : 'D'); } else if (base == long.class) { tform.append('K'); } else if (isLogical()) { tform.append('L'); } else if (isBits()) { tform.append('X'); } else if (isString()) { tform.append('A'); if (isVariableSize()) { appendSubstringConvention(tform); } } else { throw new FitsException("Invalid column data class:" + base); } return tform.toString(); } /** * Returns the TDIM header value that descrives the shape of entries in this column * * @return the TDIM header value to use, ornull
if this column is not suited for a TDIM entry for * example because it is variable-sized, or because its entries are not multidimensional. . */ String getTDIM() { if (isVariableSize()) { return null; } if (fitsShape.length < 2) { return null; } StringBuffer tdim = new StringBuffer(); char prefix = '('; for (int i = fitsShape.length - 1; i >= 0; i--) { tdim.append(prefix); tdim.append(fitsShape[i]); prefix = ','; } tdim.append(')'); return tdim.toString(); } private boolean setFitsType(char type) throws FitsException { switch (type) { case 'A': fitsBase = byte.class; base = String.class; break; case 'X': fitsBase = byte.class; base = boolean.class; break; case 'L': fitsBase = byte.class; base = boolean.class; break; case 'B': fitsBase = byte.class; base = byte.class; break; case 'I': fitsBase = short.class; base = short.class; break; case 'J': fitsBase = int.class; base = int.class; break; case 'K': fitsBase = long.class; base = long.class; break; case 'E': case 'C': fitsBase = float.class; base = float.class; break; case 'D': case 'M': fitsBase = double.class; base = double.class; break; default: return false; } return true; } } /** * The enclosing binary table's properties * * @deprecated (for internal use) no longer used, and will be removed in the future. */ protected static class SaveState { /** * Create a new saved state * * @param columns the column descriptions to save * @param heap the heap to save * * @deprecated (for internal use) no longer in use. Will remove in the future. */ public SaveState(Listcolumns, FitsHeap heap) { } } /** * Our own Logger instance, for nothing various non-critical issues. */ private static final Logger LOG = Logger.getLogger(BinaryTable.class.getName()); /** * This is the area in which variable length column data lives. */ private FitsHeap heap; /** * The heap start from the head of the HDU */ private long heapAddress; /** * (bytes) Empty space to leave after the populated heap area for future additions. */ private int heapReserve; /** * The original heap size (from the header) */ private int heapFileSize; /** * A list describing each of the columns in the table */ private List columns; /** * The number of rows in the table. */ private int nRow; /** * The length in bytes of each row. */ private int rowLen; /** * Where the data is actually stored. */ private ColumnTable> table; private FitsEncoder encoder; /** * Creates an empty binary table, which can be populated with columns / rows as desired. */ public BinaryTable() { table = new ColumnTable<>(); columns = new ArrayList<>(); heap = new FitsHeap(0); nRow = 0; rowLen = 0; } /** * Creates a binary table from an existing column table. WARNING!, as of 1.18 we no longer use the column * data extra state to carry information about an enclosing class, because it is horribly bad practice. You should * not use this constructor to create imperfect copies of binary tables. Rather, use {@link #copy()} if you want to * create a new binary table, which properly inherits ALL of the properties of an original one. As for this * constructor, you should assume that it will not use anything beyond what's available in any generic vanilla * column table. * * @param tab the column table to create the binary table from. It must be a regular column table * that contains regular data of scalar or fixed 1D arrays only (not heap pointers). * No information beyond what a generic vanilla column table provides will be used. * Column tables don't store imensions for their elements, and don't have * variable-sized entries. Thus, if the table was the used in another binary table to * store flattened multidimensional data, we'll detect that data as 1D arrays. Andm if * the table was used to store heap pointers for variable length arrays, we'll detect * these as regular int[2]
orlong[2]
values. * * @deprecated DO NOT USE -- it will be removed in the future. * * @throws FitsException if the table could not be copied and threw a {@link nom.tam.util.TableException}, which * is preserved as the cause. * * @see #copy() */ public BinaryTable(ColumnTable> tab) throws FitsException { this(); table = new ColumnTable<>(); nRow = tab.getNRows(); columns = new ArrayList<>(); for (int i = 0; i < tab.getNCols(); i++) { int n = tab.getElementSize(i); ColumnDesc c = new ColumnDesc(tab.getElementClass(i), n > 1 ? new int[] {n} : SINGLETON_SHAPE); addFlattenedColumn(tab.getColumn(i), nRow, c, true); } } /** * Creates a binary table from a given FITS header description. The table columns are initialized but no data will * be available, at least initially. Data may be loaded later (e.g. deferred read mode), provided the table is * associated to an input (usually only if this constructor is called from a {@link Fits} object reading an input). * When the table has an input configured via a {@link Fits} object, the table entries may be accessed in-situ in * the file while in deferred read mode, but operations affecting significant portions of the table (e.g. retrieving * all data via {@link #getData()} or accessing entire columns) may load the data in memory. You can also call * {@link #detach()} any time to force loading the data into memory, so that alterations after that will not be * reflected in the original file, at least not unitl {@link #rewrite()} is called explicitly. * * @param header A FITS header describing what the binary table should look like. * * @throws FitsException if the specified header is not usable for a binary table * * @deprecated (for internal use) This constructor should only be called from a {@link Fits} * object reading an input; visibility may be reduced to the package level in the * future. * * @see #isDeferred() */ public BinaryTable(Header header) throws FitsException { String ext = header.getStringValue(Standard.XTENSION, Standard.XTENSION_IMAGE); if (!ext.equalsIgnoreCase(Standard.XTENSION_BINTABLE) && !ext.equalsIgnoreCase(NonStandard.XTENSION_A3DTABLE)) { throw new FitsException( "Not a binary table header (XTENSION = " + header.getStringValue(Standard.XTENSION) + ")"); } nRow = header.getIntValue(Standard.NAXIS2); long tableSize = nRow * header.getLongValue(Standard.NAXIS1); long paramSizeL = header.getLongValue(Standard.PCOUNT); long heapOffsetL = header.getLongValue(Standard.THEAP, tableSize); // Subtract out the size of the regular table from // the heap offset. long heapSizeL = (tableSize + paramSizeL) - heapOffsetL; if (heapSizeL < 0) { throw new FitsException("Inconsistent THEAP and PCOUNT"); } if (heapSizeL > Integer.MAX_VALUE) { throw new FitsException("Heap size > 2 GB"); } if (heapSizeL == 0L) { // There is no heap. Forget the offset heapAddress = 0; } heapAddress = (int) heapOffsetL; heapFileSize = (int) heapSizeL; int nCol = header.getIntValue(Standard.TFIELDS); rowLen = 0; columns = new ArrayList<>(); for (int col = 0; col < nCol; col++) { rowLen += processCol(header, col, rowLen); } HeaderCard card = header.getCard(Standard.NAXIS1); card.setValue(rowLen); } /** * Creates a binary table from existing table data int row-major format. That is the first array index is the row * index while the second array index is the column index. * * @param rowColTable Row / column array. Scalars elements are wrapped in arrays of 1, s.t. a single *int
elements is stored asint[1]
at its *[row][col]
index. * * @throws FitsException if the argument is not a suitable representation of data in rows. * * @deprecated The constructor is ambiguous, use {@link #fromRowMajor(Object[][])} instead. You can * have a column-major array that has no scalar primitives which would also be an *Object[][]
and could be passed erroneously. */ public BinaryTable(Object[][] rowColTable) throws FitsException { this(); for (Object[] row : rowColTable) { addRow(row); } } /** * Creates a binary table from existing table data in row-major format. That is the first array index is the row * index while the second array index is the column index; * * @param table Row / column array. Scalars elements are wrapped in arrays of 1, s.t. a single *int
elements is stored asint[1]
at its *[row][col]
index. * * @return a new binary table with the data. The tables data may be partially independent from the * argument. Modifications to the table data, or that to the argument have undefined * effect on the other object. If it is important to decouple them, you can use a * {@link ArrayFuncs#deepClone(Object)} of your original data as an argument. * * @throws FitsException if the argument is not a suitable representation of FITS data in rows. * * @see #fromColumnMajor(Object[]) * * @since 1.18 */ public static BinaryTable fromRowMajor(Object[][] table) throws FitsException { BinaryTable tab = new BinaryTable(); for (Object[] row : table) { tab.addRow(row); } return tab; } /** * Create a binary table from existing data in column-major format order. * * @param columns array of columns. The data for scalar entries is a primive array. For all else, the * entry is anObject[]
array of sorts. * * @throws FitsException if the data for the columns could not be used as coulumns * * @deprecated The constructor is ambiguous, use {@link #fromColumnMajor(Object[])} instead. One could * call this method with any row-majorObject[][]
table by mistake. * * @see #defragment() */ public BinaryTable(Object[] columns) throws FitsException { this(); for (Object element : columns) { addColumn(element); } } /** * Creates a binary table from existing data in column-major format order. * * @param columns array of columns. The data for scalar entries is a primive array. For all else, the entry * is anObject[]
array of sorts. * * @return a new binary table with the data. The tables data may be partially independent from the * argument. Modifications to the table data, or that to the argument have undefined * effect on the other object. If it is important to decouple them, you can use a * {@link ArrayFuncs#deepClone(Object)} of your original data as an argument. * * @throws FitsException if the argument is not a suitable representation of FITS data in rows. * * @see #fromColumnMajor(Object[]) * * @since 1.18 */ public static BinaryTable fromColumnMajor(Object[] columns) throws FitsException { BinaryTable t = new BinaryTable(); for (Object element : columns) { t.addColumn(element); } return t; } @Override protected BinaryTable clone() { try { return (BinaryTable) super.clone(); } catch (CloneNotSupportedException e) { return null; } } /** * Returns an independent copy of the binary table. * * @return a new binary that tnat contains an exact copy of us, but is completely independent. * * @throws FitsException if the table could not be copied * * @since 1.18 */ public synchronized BinaryTable copy() throws FitsException { BinaryTable copy = clone(); if (table != null) { copy.table = table.copy(); } if (heap != null) { synchronized (copy) { copy.heap = heap.copy(); } } copy.columns = new ArrayList<>(); for (ColumnDesc c : columns) { c = c.clone(); copy.columns.add(c); } return copy; } /** * (for internal use) Discards all variable-length arrays from this table, that is all data stored on the * heap, and resets all heap descritors to (0,0). * * @since 1.19.1 */ protected synchronized void discardVLAs() { for (int col = 0; col < columns.size(); col++) { ColumnDesc c = columns.get(col); if (c.isVariableSize()) { for (int row = 0; row < nRow; row++) { table.setElement(row, col, c.hasLongPointers() ? new long[2] : new int[2]); } } } heap = new FitsHeap(0); } /** * Returns the number of bytes per regular table row * * @return the number of bytes in a regular table row. */ final int getRowBytes() { return rowLen; } /** * @deprecated (for internal use) It may become a private method in the future. * * @param table the table to create the column data. * * @throws FitsException if the data could not be created. */ public static void createColumnDataFor(BinaryTable table) throws FitsException { table.createTable(table.nRow); } /** * @deprecated (for internal use) It may be reduced to private visibility in the future. Parse the * TDIMS value. If the TDIMS value cannot be deciphered a one-d array with the size given in * arrsiz is returned. * * @param tdims The value of the TDIMSn card. * * @return An int array of the desired dimensions. Note that the order of the tdims is the inverse of the * order in the TDIMS key. */ public static int[] parseTDims(String tdims) { if (tdims == null) { return null; } // The TDIMs value should be of the form: "(i,j...)" int start = tdims.indexOf('('); if (start < 0) { return null; } int end = tdims.indexOf(')', start); if (end < 0) { end = tdims.length(); } StringTokenizer st = new StringTokenizer(tdims.substring(start + 1, end), ","); int dim = st.countTokens(); if (dim > 0) { int[] dims = new int[dim]; for (int i = dim; --i >= 0;) { dims[i] = Integer.parseInt(st.nextToken().trim()); } return dims; } return null; } /** ** Adds a column of complex values stored as the specified decimal type of components in the FITS. While you can * also use {@link #addColumn(Object)} to add complex values, that method will always add them as 64-bit * double-precision values. So, this method is provided to allow users more control over how they want their complex * data be stored. *
** The new column will be named as "Column n" (where n is the 1-based index of the column) by default, * which can be changed by {@link ColumnDesc#name(String)} after. *
* * @param o A {@link ComplexValue} or an array (possibly multi-dimensional) thereof. * @param decimalTypefloat.class
ordouble.class
(all other values default to *double.class
). * * @return the number of column in the table including the new column. * * @throws FitsException if the object contains values other than {@link ComplexValue} types or if the array is not * suitable for storing in the FITS, e.g. because it is multi-dimensional but varying in * shape / size. * * @since 1.18 * * @see #addColumn(Object) */ public int addComplexColumn(Object o, Class> decimalType) throws FitsException { int col = columns.size(); int eSize = addColumn(ArrayFuncs.complexToDecimals(o, decimalType)); ColumnDesc c = columns.get(col); c.isComplex = true; c.setLegacyShape(c.fitsShape); return eSize; } /** ** Adds a column of string values (one per row), optimized for storage size. Unlike {@link #addColumn(Object)}, * which always store strings in fixed format, this method will automatically use variable-length columns for * storing the strings if their lengths vary sufficiently to make that form of storage more efficient, or if the * array contains nulls (which may be defined later). *
** The new column will be named as "Column n" (where n is the 1-based index of the column) by default, * which can be changed by {@link ColumnDesc#name(String)} after. *
* * @param o A 1D string array, with 1 string element per table row. The array may contain *null
entries, in which case variable-length columns will be used, since * these may be defined later... * * @return the number of column in the table including the new column. * * @throws FitsException if the object contains values other than {@link ComplexValue} types or if the array is not * suitable for storing in the FITS, e.g. because it is multi-dimensional but varying in * shape / size. * * @since 1.18 * * @see #addColumn(Object) */ public int addStringColumn(String[] o) throws FitsException { checkRowCount(o); ColumnDesc c = new ColumnDesc(String.class); // Check if we should be using variable-length strings // (provided its a scalar string column with sufficiently varied strings sizes to make it worth.. int min = FitsUtil.minStringLength(o); int max = FitsUtil.maxStringLength(o); if (max - min > 2 * ElementType.forClass(c.pointerClass()).size()) { c = ColumnDesc.createForVariableSize(String.class); return addVariableSizeColumn(o, c); } c = ColumnDesc.createForStrings(max); return addFlattenedColumn(o, o.length, c, false); } /** ** Adds a column of bits. This uses much less space than if adding boolean values as logicals (the default behaviot * of {@link #addColumn(Object)}, since logicals take up 1 byte per element, whereas bits are really single bits. *
** The new column will be named as "Column n" (where n is the 1-based index of the column) by default, * which can be changed by {@link ColumnDesc#name(String)} after. *
* * @param o An any-dimensional array ofboolean
values. * * @return the number of column in the table including the new column. * * @throws IllegalArgumentException if the argument is not an array ofboolean
values. * @throws FitsException if the object is not an array ofboolean
values. * * @since 1.18 * * @see #addColumn(Object) */ public int addBitsColumn(Object o) throws FitsException { if (ArrayFuncs.getBaseClass(o) != boolean.class) { throw new IllegalArgumentException("Not an array of booleans: " + o.getClass()); } return addColumn(o, false); } /** ** Adds a new empty column to the table to the specification. This is useful when the user may want ot have more * control on how columns are configured before calling {@link #addRow(Object[])} to start populating. The new * column will be named as "Column n" (where n is the 1-based index of the column) by default, unless * already named otherwise. *
** The new column will be named as "Column n" (where n is the 1-based index of the column) by default, * which can be changed by {@link ColumnDesc#name(String)} after. *
* * @param descriptor the column descriptor * * @return the number of table columns after the addition * * @throws IllegalStateException if the table already contains data rows that prevent the addition of empty * comlumns. * * @see #addRow(Object[]) * @see ColumnDesc#name(String) */ public int addColumn(ColumnDesc descriptor) throws IllegalStateException { if (nRow != 0) { throw new IllegalStateException("Cannot add empty columns to table already containing data rows"); } descriptor.offset = rowLen; rowLen += descriptor.rowLen(); if (descriptor.name() == null) { // Set default column name; descriptor.name(TableHDU.getDefaultColumnName(columns.size())); } columns.add(descriptor); return columns.size(); } /** * Converts a boxed table entry to an array. * * @param o a boxed table entry or array of some kind * * @return an array object that wrap non-array arguments * * @throws FitsException If the argument is not a valid FITS object */ private static Object entryToColumnArray(Object o) throws FitsException { o = boxedToArray(o); if (o.getClass().isArray()) { int[] dim = ArrayFuncs.getDimensions(o); if (dim.length == 1 && dim[0] == 1) { return o; } } Object[] array = (Object[]) Array.newInstance(o.getClass(), 1); array[0] = o; return array; } /** ** Adds a new column with the specified data array, with some default mappings. This method will always use * double-precision representation for {@link ComplexValue}-based data, and will represent
*boolean
* based array data as one-byte-per element FITS logical values (for back compatibility). It will also store strings * as fixed sized (sized for the longest string element contained). ** The new column will be named as "Column n" (where n is the 1-based index of the column) by default, * which can be changed by {@link ColumnDesc#name(String)} after. *
** If you want other complex-valued representations use {@link #addComplexColumn(Object, Class)} instead, and if you * want to pack
*boolean
-based data more efficiently (using up to 8 times less space), use * {@link #addBitsColumn(Object)} instead, or else convert the column to bits afterwards using * {@link #convertToBits(int)}. And, if you want to allow storing strings more effiently in variable-length columns, * you should use {@link #addStringColumn(String[])} instead. ** As of 1.18, the argument can be a boxed primitive for a coulmn containing a single scalar-valued entry (row). *
* * @see #addVariableSizeColumn(Object) * @see #addComplexColumn(Object, Class) * @see #addBitsColumn(Object) * @see #convertToBits(int) * @see #addStringColumn(String[]) * @see ColumnDesc#name(String) */ @Override public int addColumn(Object o) throws FitsException { return addColumn(o, true); } private int checkRowCount(Object o) throws FitsException { if (!o.getClass().isArray()) { throw new TableException("Not an array: " + o.getClass().getName()); } int rows = Array.getLength(o); if (columns.size() != 0 && rows != nRow) { throw new TableException("Mismatched number of rows: " + rows + ", expected " + nRow); } return rows; } /** * Like {@link #addColumn(Object)}, but allows specifying whether we use back compatible mode. This mainly just * affects howboolean
arrays are stored (as logical bytes in compatibility mode, or as packed bits * otherwise). * * @param Whether to add the column in a back compatibility mode with versions prior to 1.18. Iftrue
*boolean
arrays will stored as logical bytes, otherwise as packed bits. */ private int addColumn(Object o, boolean compat) throws FitsException { o = boxedToArray(o); int rows = checkRowCount(o); ColumnDesc c = new ColumnDesc(ArrayFuncs.getBaseClass(o)); if (ArrayFuncs.getBaseClass(o) == ComplexValue.class) { o = ArrayFuncs.complexToDecimals(o, double.class); c.isComplex = true; } try { int[] dim = ArrayFuncs.checkRegularArray(o, c.isNullAllowed()); if (c.isString()) { c.setStringLength(FitsUtil.maxStringLength(o)); } if (c.isComplex) { // Drop the railing 2 dimension, keep only outer dims... dim = Arrays.copyOf(dim, dim.length - 1); o = ArrayFuncs.flatten(o); } if (dim.length <= 1) { c.setSingleton(); } else { int[] shape = new int[dim.length - 1]; System.arraycopy(dim, 1, shape, 0, shape.length); c.setLegacyShape(shape); o = ArrayFuncs.flatten(o); } } catch (IllegalArgumentException e) { c.setVariableSize(false); return addVariableSizeColumn(o, c); } // getBaseClass() prevents heterogeneous columns, so no need to catch ClassCastException here. return addFlattenedColumn(o, rows, c, compat); } /** ** Adds a new variable-length data column, populating it with the specified data object. Unlike * {@link #addColumn(Object)} which will use fixed-size data storage provided the data allows it, this method forces * the use of variable-sized storage regardless of the data layout -- for example to accommodate addiing rows / * elements of different sized at a later time. *
** The new column will be named as "Column n" (where n is the 1-based index of the column) by default, * which can be changed by {@link ColumnDesc#name(String)} after. *
* * @param o An array containing one entry per row. Multi-dimensional entries will be flattened to 1D * for storage on the heap. * * @return the number of table columns after the addition. * * @throws FitsException if the column could not be created as requested. * * @see #addColumn(Object) * @see #addColumn(ColumnDesc) * @see ColumnDesc#createForVariableSize(Class) * @see ColumnDesc#isVariableSize() * * @since 1.18 */ public int addVariableSizeColumn(Object o) throws FitsException { Class> base = ArrayFuncs.getBaseClass(o); ColumnDesc c = ColumnDesc.createForVariableSize(base); return addVariableSizeColumn(o, c); } /** * Adds a new column with data directly, without performing any checks on the data. This should only be use * internally, after ansuring the data integrity and suitability for this table. * * @param o the column data, whose integrity was verified previously * @param rows the number of rows the data contains (in flattened form) * @param c the new column's descriptor * * @return the number of table columns after the addition * * @throws FitsException if the data is not the right type or format for internal storage. */ private int addDirectColumn(Object o, int rows, ColumnDesc c) throws FitsException { c.offset = rowLen; rowLen += c.rowLen(); // Load any deferred data (we will not be able to do that once we alter the column structure) ensureData(); // Set the default column name c.name(TableHDU.getDefaultColumnName(columns.size())); table.addColumn(o, c.getTableBaseCount()); columns.add(c); if (nRow == 0) { // Set the table row count to match first colum nRow = rows; } return columns.size(); } private int addVariableSizeColumn(Object o, ColumnDesc c) throws FitsException { checkRowCount(o); Object[] array = (Object[]) o; o = Array.newInstance(c.pointerClass(), array.length * 2); for (int i = 0; i < array.length; i++) { boolean multi = c.isComplex() ? array[i] instanceof Object[][] : array[i] instanceof Object[]; if (multi) { boolean canBeComplex = false; if (c.getFitsBase() == float.class || c.getFitsBase() == double.class) { int[] dim = ArrayFuncs.getDimensions(array[i]); if (dim[dim.length - 1] == 2) { canBeComplex = true; } } if (!canBeComplex && !c.warnedFlatten) { LOG.warning("Table entries of " + array[i].getClass() + " will be stored as 1D arrays in variable-length columns. " + "Array shape(s) and intermittent null subarrays (if any) will be lost."); c.warnedFlatten = true; } } Object p = putOnHeap(c, array[i], null); System.arraycopy(p, 0, o, 2 * i, 2); } return addDirectColumn(o, array.length, c); } /** * Add a column where the data is already flattened. * * @param o The new column data. This should be a one-dimensional primitive array. * @param dims The dimensions of an element in the column, or null for singleton (scalar) columns * * @return the new column size * * @throws FitsException if the array could not be flattened * * @deprecated (for internal use) No longer used, will be removed in the future */ public int addFlattenedColumn(Object o, int... dims) throws FitsException { ColumnDesc c = new ColumnDesc(ArrayFuncs.getBaseClass(o)); try { ArrayFuncs.checkRegularArray(o, c.isNullAllowed()); } catch (IllegalArgumentException e) { throw new FitsException("Irregular array: " + o.getClass() + ": " + e.getMessage(), e); } if (c.isString()) { c.setStringLength(FitsUtil.maxStringLength(o)); } int n = 1; c.setLegacyShape(dims); for (int dim : dims) { n *= dim; } int rows = Array.getLength(o) / n; return addFlattenedColumn(o, rows, c, true); } /** * Checks that a flattened column has a compatible size for storing in a fixed-width column. It will also log a * warning if the storage size of the object is zero. * * @param c the column descriptor * @param o the column data * * @throws FitsException if the data is not the right size for the column */ private void checkFlattenedColumnSize(ColumnDesc c, Object o) throws FitsException { if (c.getTableBaseCount() == 0) { LOG.warning("Elements of column + " + columns.size() + " have zero storage size."); } else if (columns.size() > 0) { // Check that the number of rows is consistent. int l = Array.getLength(o); if (nRow > 0 && l != nRow * c.getTableBaseCount()) { throw new TableException("Mismatched element count " + l + ", expected " + (nRow * c.getTableBaseCount())); } } } /** * This function is needed since we had made addFlattenedColumn public so in principle a user might have called it * directly. * * @param o The new column data. This should be a one-dimensional primitive array. * @param c The column description * * @return the new column size * * @throws FitsException if the data type, format, or element count is inconsistent with this table. */ private int addFlattenedColumn(Object o, int rows, ColumnDesc c, boolean compat) throws FitsException { // For back compatibility this method will add boolean values as logicals always... if (compat) { c.isBits = false; } if (c.isBits) { // Special handling for bits, which have to be segmented into bytes... boolean[] bits = (boolean[]) o; o = FitsUtil.bitsToBytes(bits, bits.length / rows); } else { o = javaToFits1D(c, o); } checkFlattenedColumnSize(c, o); return addDirectColumn(o, rows, c); } /** ** Adds a row to the table. If this is the first row in a new table, fixed-length columns will be created from the * data type automatically. If you want more control over the column formats, you may want to specify columns * beforehand such as: *
* ** BinaryTable table = new BinaryTable(); * * // A column containing 64-bit floating point scalar values, 1 per row... * table.addColumn(ColumnDesc.createForScalars(double.class)); * * // A column containing 5x4 arrays of single-precision complex values... * table.addColumn(ColumnDesc.createForArrays(ComplexValue.Float.class, 5, 4) * * // A column containing Strings of variable length using 32-bit heap pointers... * table.addColumn(ColumnDesc.creatForVariableStrings(false); *** For scalar columns of primitive types, the argument may be the corresponding java boxed type (new style), or a * primitive array of 1 (old style). Thus, you can write either: *
* ** table.addRow(1, 3.14159265); *** or, *
* ** table.addRow(new Object[] { new int[] {1}, new double[] {3.14159265} }; ** * @see #addColumn(ColumnDesc) */ @Override public int addRow(Object[] o) throws FitsException { ensureData(); if (columns.isEmpty()) { for (Object element : o) { if (element == null) { throw new TableException("Prototype row may not contain null"); } addColumn(entryToColumnArray(element)); } } else if (o.length != columns.size()) { throw new TableException("Mismatched row size: " + o.length + ", expected " + columns.size()); } else { Object[] flatRow = new Object[getNCols()]; for (int i = 0; i < flatRow.length; i++) { ColumnDesc c = columns.get(i); flatRow[i] = c.isVariableSize() ? putOnHeap(c, o[i], null) : javaToFits1D(c, ArrayFuncs.flatten(o[i])); } table.addRow(flatRow); nRow++; } return nRow; } @Override public void deleteColumns(int start, int len) throws FitsException { ensureData(); table.deleteColumns(start, len); ArrayListremain = new ArrayList<>(columns.size() - len); rowLen = 0; for (int i = 0; i < columns.size(); i++) { if (i < start || i >= start + len) { ColumnDesc c = columns.get(i); c.offset = rowLen; rowLen += c.rowLen(); remain.add(c); } } columns = remain; } @Override public void deleteRows(int row, int len) throws FitsException { ensureData(); table.deleteRows(row, len); nRow -= len; } /** * Returns the Java type of elements returned or expected by the older srray-based access methods. It can be * confusing, because: * *
* * @return the types in the table, not the underlying types (e.g., for varying length arrays or booleans). * * @deprecated (for internal use) Ambiguous, use {@link ColumnDesc#getElementClass()} instead. Will remove in * the future. */ public Class>[] getBases() { return table.getBases(); } /** *- Columns with variable sized entries report
*int.class
orlong.class
regardless of * data type.- Regular logical and bit columns bith report
*boolean.class
.- Regular complex valued columns report
*float.class
ordouble.class
.* Returns the data for a particular column in as an array of elements. See {@link #addColumn(Object)} for more * information about the format of data elements in general. *
* * @param col The zero-based column index. * * @return an array of primitives (for scalar columns), or else anObject[]
array, or * possiblynull
* * @throws FitsException if the table could not be accessed * * @see #setColumn(int, Object) * @see #getElement(int, int) * @see #getNCols() */ @Override public Object getColumn(int col) throws FitsException { ColumnDesc c = columns.get(col); if (!c.isVariableSize() && c.fitsDimension() == 0 && !c.isComplex()) { return getFlattenedColumn(col); } ensureData(); Object[] data = null; for (int i = 0; i < nRow; i++) { Object e = getElement(i, col); if (data == null) { data = (Object[]) Array.newInstance(e.getClass(), nRow); } data[i] = e; } return data; } /** * Returns the Java index of the first column by the specified name. * * @param name the name of the column (case sensitive). * * @return The column index, or else -1 if this table does not contain a column by the specified name. * * @see #getDescriptor(String) * @see ColumnDesc#name(String) * * @since 1.20 * * @author Attila Kovacs */ public int indexOf(String name) { for (int col = 0; col < columns.size(); col++) { if (name.equals(getDescriptor(col).name())) { return col; } } return -1; } @Override protected ColumnTable> getCurrentData() { return table; } @Override public ColumnTable> getData() throws FitsException { return (ColumnTable>) super.getData(); } /** * Returns the dimensions of elements in each column. * * @return an array of arrays with the dimensions of each column's data. * * @see ColumnDesc#getDimens() * * @deprecated (for internal use) Use {@link ColumnDesc#getEntryShape()} to access the shape of Java elements * individually for columns instead. Not useful to users since it returns the dimensions of the * primitive storage types, which is not always the dimension of elements on the Java side (notably * for string entries). */ public int[][] getDimens() { int[][] dimens = new int[columns.size()][]; for (int i = 0; i < dimens.length; i++) { dimens[i] = columns.get(i).getDimens(); } return dimens; } /** * @deprecated (for internal use) It may be private in the future. * * @return An array with flattened data, in which each column's data is represented by a 1D array * * @throws FitsException if the reading of the data failed. */ public Object[] getFlatColumns() throws FitsException { ensureData(); return table.getColumns(); } /** * @deprecated (for internal use) It may be reduced to private visibility in the future. * * @return column in flattened format. This is sometimes useful for fixed-sized columns. * Variable-sized columns will still return anObject[]
array in which * each entry is the variable-length data for a row. * * @param col the column to flatten * * @throws FitsException if the column could not be flattened */ public Object getFlattenedColumn(int col) throws FitsException { if (!validColumn(col)) { throw new TableException("Invalid column index " + col + " in table of " + getNCols() + " columns"); } ColumnDesc c = columns.get(col); if (c.isVariableSize()) { throw new TableException("Cannot flatten variable-sized column data"); } ensureData(); if (c.isBits()) { boolean[] bits = new boolean[nRow * c.fitsCount]; for (int i = 0; i < nRow; i++) { boolean[] seg = (boolean[]) fitsToJava1D(c, table.getElement(i, col), c.fitsCount, false); System.arraycopy(seg, 0, bits, i * c.fitsCount, c.fitsCount); } return bits; } return fitsToJava1D(c, table.getColumn(col), 0, false); } /** ** Reserves space for future addition of rows at the end of the regular table. In effect, this pushes the heap to * start at an offset value, leaving a gap between the main table and the heap in the FITS file. If your table * contains variable-length data columns, you may also want to reserve extra heap space for these via * {@link #reserveHeapSpace(int)}. *
** Note, that (C)FITSIO, as of version 4.4.0, has no proper support for offset heaps, and so you may want to be * careful using this function as the resulting FITS files, while standard, may not be readable by other tools due * to their own lack of support. Note, however, that you may also use this function to undo an offset heap with an * argument <=0; *
* * @param rows The number of future rows fow which space should be reserved (relative to the current table size) * for future additions, or <=0 to ensure that the heap always follows immediately after the * main table, e.g. for better (C)FITSIO interoperability. * * @see #reserveHeapSpace(int) * * @since 1.19.1 * * @author Attila Kovacs */ public void reserveRowSpace(int rows) { heapAddress = rows > 0 ? getRegularTableSize() + (long) rows * getRowBytes() : 0; } /** * Reserves space in the file at the end of the heap for future heap growth (e.g. different/longer or new VLA * entries). You may generally want to call this along with {@link #reserveRowSpace(int)} if yuor table contains * variable-length columns, to ensure storage for future data in these. You may call with <=0 to discards any * previously reserved space. * * @param bytes The number of bytes of unused space to reserve at the end of the heap, e.g. for future * modifications or additions, when writing the data to file. * * @see #reserveRowSpace(int) * * @since 1.19.1 * * @author Attila Kovacs */ public void reserveHeapSpace(int bytes) { heapReserve = Math.max(0, bytes); } /** * Returns the address of the heap from the star of the HDU in the file. * * @return (bytes) the start of the heap area from the beginning of the HDU. */ final long getHeapAddress() { long tableSize = getRegularTableSize(); return heapAddress > tableSize ? heapAddress : tableSize; } /** * Returns the offset from the end of the main table * * @return the offset to the heap */ final long getHeapOffset() { return getHeapAddress() - getRegularTableSize(); } /** * It returns the heap size for storing in the FITS, which is the larger of the actual space occupied by the current * heap, or the original heap size based on the header when the HDU was read from an input. In the former case it * will also include heap space reserved for future additions. * * @return (byte) the size of the heap in the FITS file. * * @see #compact() * @see #reserveHeapSpace(int) */ private int getHeapSize() { if (heap != null && heap.size() + heapReserve > heapFileSize) { return heap.size() + heapReserve; } return heapFileSize; } /** * @return the size of the heap -- including the offset from the end of the table data, and reserved space after. */ synchronized long getParameterSize() { return getHeapOffset() + getHeapSize(); } /** * Returns an empty row for the table. Such model rows are useful when low-level reading binary tables from an input * row-by-row. You can simply all {@link nom.tam.util.ArrayDataInput#readArrayFully(Object)} to populate it with * data from a stream. You may also use model rows to add additional rows to an existing table. * * @return a row that may be used for direct i/o to the table. * * @deprecated (for internal use) Use {@link #getElement(int, int)} instead for low-level reading of tables * in deferred mode. Not recommended for uses because it requires a deep understanding of how data * (especially varialbe length columns) are represented in the FITS. Will reduce visibility to * private in the future. */ public Object[] getModelRow() { Object[] modelRow = new Object[columns.size()]; for (int i = 0; i < modelRow.length; i++) { ColumnDesc c = columns.get(i); if (c.fitsDimension() < 2) { modelRow[i] = Array.newInstance(c.getTableBase(), c.getTableBaseCount()); } else { modelRow[i] = Array.newInstance(c.getTableBase(), c.fitsShape); } } return modelRow; } @Override public int getNCols() { return columns.size(); } @Override public int getNRows() { return nRow; } /** * Reads a regular table element in the main table from the input. This method should never be called unless we have * a random-accessible input associated, which is a requirement for deferred read mode. * * @param o The array element to populate * @param c the column descriptor * @param row the zero-based row index of the element * * @throws IOException If there was an I/O error accessing the input * @throws FitsException If there was some other error */ private synchronized void readTableElement(Object o, ColumnDesc c, int row) throws IOException, FitsException { @SuppressWarnings("resource") RandomAccess in = getRandomAccessInput(); in.position(getFileOffset() + row * (long) rowLen + c.offset); if (c.isLogical()) { in.readArrayFully(o); } else { in.readImage(o); } } /** * Returns an unprocessed element from the table as a 1D array of the elements that are stored in the regular table * data, whithout reslving heap references. That is this call will return flattened versions of multidimensional * arrays, and will return only the heap locator (offset and size) for variable-sized columns. * * @return a particular element from the table but do no processing of this element (e.g., * dimension conversion or extraction of variable length array elements/) * * @param row The row of the element. * @param col The column of the element. * * @deprecated (for internal use) Will reduce visibility in the future. * * @throws FitsException if the operation failed */ public Object getRawElement(int row, int col) throws FitsException { if (!validRow(row) || !validColumn(col)) { throw new TableException("No such element (" + row + "," + col + ")"); } if (table == null) { try { ColumnDesc c = columns.get(col); Object e = c.newInstance(1); readTableElement(e, c, row); return e; } catch (IOException e) { throw new FitsException("Error reading from input: " + e.getMessage(), e); } } ensureData(); return table.getElement(row, col); } /** * Returns a table element as a Java array. Consider using the more Java-friendly {@link #get(int, int)} or one of * the scalar access methods with implicit type conversion support. * * @see #get(int, int) * @see #getLogical(int, int) * @see #getNumber(int, int) * @see #getLong(int, int) * @see #getDouble(int, int) * @see #getString(int, int) */ @Override public Object getElement(int row, int col) throws FitsException { return getElement(row, col, false); } /** * Returns a a table entry, with control over how FITS logical values are to be handled. * * @param row zero-based row index * @param col zero-based column index * @param isEnhanced Whether logicals should be returned as {@link Boolean} (rather thanboolean
) * and complex values as {@link ComplexValue} (rather thanfloat[2]
or *double[2]
), or arrays thereof. Methods prior to 1.18 should set this to *false
for back compatible behavior. * * @return The entry as a primitive array, or {@link String}, {@link Boolean} or {@link ComplexValue}, * or arrays thereof. * * @throws FitsException If the requested element could not be accessed. */ private Object getElement(int row, int col, boolean isEnhanced) throws FitsException { if (!validRow(row) || !validColumn(col)) { throw new TableException("No such element (" + row + "," + col + ")"); } ColumnDesc c = columns.get(col); Object o = getRawElement(row, col); if (c.isVariableSize()) { return getFromHeap(c, o, isEnhanced); } o = fitsToJava1D(c, o, c.isBits() ? c.fitsCount : 0, isEnhanced); if (c.legacyShape.length > 1) { return ArrayFuncs.curl(o, c.legacyShape); } return o; } /** * Returns a table element as an array of the FITS storage type. Similar to the original * {@link #getElement(int, int)}, except that FITS logicals are returned as arrays ofBoolean
(rather * thanboolean
), bits are returned as arrays ofboolean
, and complex values are returned * as arrays of {@link ComplexValue} rather than arrays ofdouble[2]
orfloat[2]
. * Singleton (scalar) table elements are not boxed to an enclosing Java type (unlike {@link #get(int, int)}), an * instead returned as arrays of just one element. For example, a single logical as aBoolean[1]
, a * single float as afloat[1]
or a single double-precision complex value as *ComplexValue[1]
. * * @param row zero-based row index * @param col zero-based column index * * @return The table entry as an array of the stored Java type, without applying any type or quantization * conversions. * * @see #getArrayElementAs(int, int, Class) * @see #get(int, int) * * @since 1.20 */ public Object getArrayElement(int row, int col) { return getElement(row, col, true); } /** ** Returns a numerical table element as an array of a specific underlying other numerical type. Similar * {@link #getArrayElement(int, int)} except that table entries are converted to the specified array type before * returning. If an integer-decimal conversion is involved, it will be performed through the column's quantizer (if * any) or else via a simple rounding as necessary. *
** For example, if you have an
* * @param row zero-based row index * @param col zero-based column index * @param asType The desired underlying type, a primitive class or a {@link ComplexValue} type * for appropriate numerical arrays (with a trailing Java dimension of 2 for * the real/imaginary pairs). * * @return An array of the desired type (e.g.short
-type column, and you want is an array ofdouble
* values that are represented by the 16-bit integers, then the conversion will use the column's quantizer scaling * and offset before returning the result either as an array of doubles, and the designatedshort
* blanking values will be converted to NaNs. *double[][]
if *asType
isdouble.class
and the column contains 2D * arrays of some numerical type). * * @throws IllegalArgumentException if the numerical conversion is not possible for the given column type or if the * type argument is not a supported numerical primitive or {@link ComplexValue} * type. * * @see #getArrayElement(int, int) * * @since 1.20 */ public Object getArrayElementAs(int row, int col, Class> asType) throws IllegalArgumentException { ColumnDesc c = getDescriptor(col); Object e = getElement(row, col, true); return asType.isAssignableFrom(c.getFitsBase()) ? e : ArrayFuncs.convertArray(e, asType, c.getQuantizer()); } /** ** Returns a table element using the usual Java boxing for primitive scalar (singleton) entries, or packaging * complex values as {@link ComplexValue}, or as appropriate primitive or object arrays. FITS string columns return * {@link String} values. Logical (
*boolean
columns will return a {@link Boolean}, which may be *null
if undefined (as per the FITS standard). Multibit FITS bits colums return arrays of *boolean
. ** As opposed to {@link #getElement(int, int)} scalar (singleton) values are not wrapped into primitive arrays, but * return either a singular object, such as a ({@link String}, or a {@link ComplexValue}, or a boxed Java primitive. * Thus, columns containing single
*short
entries will return the selected element as a {@link Short}, * or columns containing singledouble
values will return the element as a {@link Double} and so on. ** Array columns will return the expected arrays of primitive values, or arrays of one of the mentioned types. Note * however, that logical arrays are returned as arrays of {@link Boolean}, e.g.
*Boolean[][]
, not *boolean[][]
. This is because FITS allowsnull
values for logicals beyond* true
andfalse
, which is reproduced by the boxed type, but not by the primitive type. FITS * columns of bits (generally preferrably to logicals if support fornull
values is not required) will * return arrays ofboolean
. ** Columns containing multidimensional arrays, will return the expected multidimensional array of the above * mentioned types for the FITS storage type. You can then convert numerical arrays to other types as required for * your application via {@link ArrayFuncs#convertArray(Object, Class, Quantizer)}, including any appropriate * quantization for the colummn (see {@link ColumnDesc#getQuantizer()}). *
* * @param row the zero-based row index * @param col the zero-based column index * * @return the element, either as a Java boxed type (for scalar entries), a singular Java Object, or * as a (possibly multi-dimensional) array of {@link String}, {@link Boolean}, * {@link ComplexValue}, or primitives. * * @throws FitsException if the element could not be obtained * * @see #getNumber(int, int) * @see #getLogical(int, int) * @see #getString(int, int) * @see #getArrayElementAs(int, int, Class) * @see #set(int, int, Object) * * @since 1.18 */ public Object get(int row, int col) throws FitsException { ColumnDesc c = columns.get(col); Object e = getElement(row, col, true); return (c.isSingleton() && e.getClass().isArray()) ? Array.get(e, 0) : e; } /** * Returns the numerical value, if possible, for scalar elements. Scalar numerical columns return the boxed type of * their primitive type. Thus, a column oflong
values will return {@link Long}, whereas a column of *float
values will return a {@link Float}. Logical columns will return 1 iftrue
or 0 if *false
, ornull
if undefined. Array columns and other column types will throw an * exception. * * @param row the zero-based row index * @param col the zero-based column index * * @return the number value of the specified scalar entry * * @throws FitsException if the element could not be obtained * @throws ClassCastException if the specified column in not a numerical scalar type. * @throws NumberFormatException if the it's a string column but the entry does not seem to be a number * * @see #getDouble(int, int) * @see #getLong(int, int) * @see #get(int, int) * * @since 1.18 */ public final Number getNumber(int row, int col) throws FitsException, ClassCastException, NumberFormatException { Object o = get(row, col); if (o instanceof String) { try { return Long.parseLong((String) o); } catch (NumberFormatException e) { return Double.parseDouble((String) o); } } if (o instanceof Boolean) { return ((Boolean) o) ? 1 : 0; } return (Number) o; } /** ** Returns the decimal value, if possible, of a scalar table entry. See {@link #getNumber(int, int)} for more * information on the conversion process. *
** Since version 1.20, if the column has a quantizer and stores integer elements, the conversion to double-precision * will account for the quantization of the column, if any, and will return NaN if the stored integer is the * designated blanking value (if any). To bypass quantization, you can use {@link #getNumber(int, int)} instead * followed by {@link Number#doubleValue()} to to get the stored integer values as a double. *
* * @param row the zero-based row index * @param col the zero-based column index * * @return the number value of the specified scalar entry * * @throws FitsException if the element could not be obtained * @throws ClassCastException if the specified column in not a numerical scalar type. * * @see #getNumber(int, int) * @see #getLong(int, int) * @see #get(int, int) * @see ColumnDesc#getQuantizer() * * @since 1.18 */ public final double getDouble(int row, int col) throws FitsException, ClassCastException { Number n = getNumber(row, col); if (!(n instanceof Float || n instanceof Double)) { Quantizer q = getDescriptor(col).getQuantizer(); if (q != null) { return q.toDouble(n.longValue()); } } return n == null ? Double.NaN : n.doubleValue(); } /** ** Returns a 64-bit integer value, if possible, of a scalar table entry. Boolean columns will return 1 if *
*true
or 0 iffalse
, or throw a {@link NullPointerException} if undefined. See * {@link #getNumber(int, int)} for more information on the conversion process of the stored data element. ** Additionally, since version 1.20, if the column has a quantizer and stores floating-point elements, the * conversion to integer will include the quantization, and NaN values will be converted to the designated integer * blanking values. To bypass quantization, you can use {@link #getNumber(int, int)} instead followed by * {@link Number#longValue()} to to get the stored floating point values rounded directly to a long. *
* * @param row the zero-based row index * @param col the zero-based column index * * @return the 64-bit integer number value of the specified scalar table entry. * * @throws FitsException if the element could not be obtained * @throws ClassCastException if the specified column in not a numerical scalar type. * @throws IllegalStateException if the column contains a undefined (blanking value), such as a {@link Double#NaN} * when no quantizer is set for the column, or a {@link Boolean}null
* value. * * @see #getNumber(int, int) * @see #getDouble(int, int) * @see #get(int, int) * * @since 1.18 */ public final long getLong(int row, int col) throws FitsException, ClassCastException, IllegalStateException { Number n = getNumber(row, col); if (n instanceof Float || n instanceof Double) { Quantizer q = getDescriptor(col).getQuantizer(); if (q != null) { return q.toLong(n.doubleValue()); } } if (Double.isNaN(n.doubleValue())) { throw new IllegalStateException("Cannot convert NaN to long without Quantizer"); } return n.longValue(); } /** * Returns the boolean value, if possible, for scalar elements. It will will returntrue
, or *false
, ornull
if undefined. Numerical columns will returnnull
if the * corresponding decimal value is NaN, orfalse
if the value is 0, or elsetrue
for all * non-zero values (just like in C). * * @param row the zero-based row index * @param col the zero-based column index * * @return the boolean value of the specified scalar entry, ornull
if undefined. * * @throws ClassCastException if the specified column in not a scalar boolean type. * @throws FitsException if the element could not be obtained * * @see #get(int, int) * * @since 1.18 */ @SuppressFBWarnings(value = "NP_BOOLEAN_RETURN_NULL", justification = "null has specific meaning here") public final Boolean getLogical(int row, int col) throws FitsException, ClassCastException { Object o = get(row, col); if (o == null) { return null; } if (o instanceof Number) { Number n = (Number) o; if (Double.isNaN(n.doubleValue())) { return null; } return n.longValue() != 0; } if (o instanceof Character) { char c = (Character) o; if (c == 'T' || c == 't' || c == '1') { return true; } if (c == 'F' || c == 'f' || c == '0') { return false; } return null; } if (o instanceof String) { return FitsUtil.parseLogical((String) o); } return (Boolean) o; } /** * Returns the string value, if possible, for scalar elements. All scalar columns will return the string * representation of their values, whilebyte[]
andchar[]
are converted to appropriate * strings. * * @param row the zero-based row index * @param col the zero-based column index * * @return the string representatiof the specified table entry * * @throws ClassCastException if the specified column contains array elements other thanbyte[]
or *char[]
* @throws FitsException if the element could not be obtained * * @see #get(int, int) * * @since 1.18 */ public final String getString(int row, int col) throws FitsException, ClassCastException { ColumnDesc c = columns.get(col); Object value = get(row, col); if (value == null) { return "null"; } if (!value.getClass().isArray()) { return value.toString(); } if (c.fitsDimension() > 1) { throw new ClassCastException("Cannot convert multi-dimensional array element to String"); } if (value instanceof char[]) { return String.valueOf((char[]) value).trim(); } if (value instanceof byte[]) { return AsciiFuncs.asciiString((byte[]) value).trim(); } throw new ClassCastException("Cannot convert " + value.getClass().getName() + " to String."); } @Override public Object[] getRow(int row) throws FitsException { if (!validRow(row)) { throw new TableException("Invalid row index " + row + " in table of " + getNRows() + " rows"); } Object[] data = new Object[columns.size()]; for (int col = 0; col < data.length; col++) { data[col] = getElement(row, col); } return data; } /** * Returns the flattened (1D) size of elements in each column of this table. As of 1.18, this method returns a copy * ot the array used internally, which is safe to modify. * * @return an array with the byte sizes of each column * * @deprecated (for internal use) Use {@link ColumnDesc#getElementCount()} instead. This one returns the * number of elements in the FITS representation, not in the java representation. For example, for * {@link String} entries, this returns the number of bytes stored, not the number of strings. * Similarly, for complex values it returns the number of components not the number of values. */ public int[] getSizes() { int[] sizes = new int[columns.size()]; for (int i = 0; i < sizes.length; i++) { sizes[i] = columns.get(i).getTableBaseCount(); } return sizes; } /** * Returns the size of the regular table data, before the heap area. * * @return the size of the regular table in bytes */ private long getRegularTableSize() { return (long) nRow * rowLen; } @Override protected long getTrueSize() { return getRegularTableSize() + getParameterSize(); } /** * Get the characters describing the base classes of the columns. As of 1.18, this method returns a copy ot the * array used internally, which is safe to modify. * * @return An array of type characters (Java array types), one for each column. * * @deprecated (for internal use) Use {@link ColumnDesc#getElementClass()} instead. Not very useful to users * since this returns the FITS primitive storage type for the data column. */ public char[] getTypes() { char[] types = new char[columns.size()]; for (int i = 0; i < columns.size(); i++) { types[i] = ElementType.forClass(columns.get(i).getTableBase()).type(); } return types; } @Override public void setColumn(int col, Object o) throws FitsException { ColumnDesc c = columns.get(col); if (c.isVariableSize()) { Object[] array = (Object[]) o; for (int i = 0; i < nRow; i++) { Object p = putOnHeap(c, ArrayFuncs.flatten(array[i]), getRawElement(i, col)); setTableElement(i, col, p); } } else { setFlattenedColumn(col, o); } } /** * Writes an element directly into the random accessible FITS file. Note, this call will not modify the table in * memory (if loaded). This method should never be called unless we have a valid encoder object that can handle the * writing, which is a requirement for deferred read mode. * * @param row the zero-based row index * @param col the zero-based column index * @param array an array object containing primitive types, in FITS storage format. It may be * multi-dimensional. * * @throws IOException the there was an error writing to the FITS output * * @see #setTableElement(int, int, Object) */ @SuppressWarnings("resource") private void writeTableElement(int row, int col, Object array) throws IOException { ColumnDesc c = columns.get(col); getRandomAccessInput().position(getFileOffset() + row * (long) rowLen + c.offset); encoder.writeArray(array); } /** * Sets a table element to an array in the FITS storage format. If the data is in deferred mode it will write the * table entry directly into the file. Otherwise it will update the table entry in memory. For variable sized * column, the heap will always be updated in memory, so you may want to call {@link #rewrite()} when done updating * all entries. * * @param row the zero-based row index * @param col the zero-based column index * @param o an array object containing primitive types, in FITS storage format. It may be * multi-dimensional. * * @throws FitsException if the array is invalid for the given column, or if the table could not be accessed in the * file / input. * * @see #setTableElement(int, int, Object) * @see #getRawElement(int, int) */ private void setTableElement(int row, int col, Object o) throws FitsException { if (table == null) { try { writeTableElement(row, col, o); } catch (IOException e) { throw new FitsException(e.getMessage(), e); } } else { ensureData(); table.setElement(row, col, o); } } /** * Consider using the more Java-friendly {@link #set(int, int, Object)} with implicit scalar type conversions. * * @see #set(int, int, Object) */ @Override public void setElement(int row, int col, Object o) throws FitsException { ColumnDesc c = columns.get(col); o = c.isVariableSize() ? putOnHeap(c, o, getRawElement(row, col)) : javaToFits1D(c, ArrayFuncs.flatten(o)); setTableElement(row, col, o); } /** ** The Swiss-army knife of setting table entries, including Java boxing, and with some support for automatic type * conversions. The argument may be one of the following type: *
**
*- Scalar values -- any Java primitive with its boxed type, such as a {@link Double}, or a * {@link Character}.
*- A single {@link String} or {@link ComplexValue} object. *
- An array (including multidimensional) of primitive types, or that of {@link Boolean}, {@link ComplexValue}, * or {@link String}.
** For array-type columns the argument needs to match the column type exactly. However, you may call * {@link ArrayFuncs#convertArray(Object, Class, Quantizer)} prior to setting values to convert arrays to the * desired numerical types, including the quantization that is appropriate for the column (see * {@link ColumnDesc#getQuantizer()}). *
** For scalar (single element) columns, automatic type conversions may apply, to make setting scalar columns more * flexible: *
**
*- Any numerical column can take any {@link Number} value. The conversion is as if an explicit Java cast were * applied. For example, if setting a
*double
value for a column of singleshort
values it * as if a(short)
cast were applied to the value.- Numerical colums can also take {@link Boolean} values which set the entry to 1, or 0, or to * {@link Double#isNaN()} (or the equivalent integer minimum value) if the argument is
*null
. Numerical * columns can also set {@link String} values, by parsing the string according to the numerical type of the * column.- Logical columns can set {@link Boolean} values, including
*null
values, but also any * {@link Number} type. In case of numbers, zero values map tofalse
while definite non-zero values map * totrue
. {@link Double#isNaN()} maps to anull
(or undefined) entry. Loginal columns * can be also set to the {@link String} values of 'true' or 'false', or to a {@link Character} of 'T'/'F' (or * equivalently '1'/'0') and 0 (undefined)- Singular string columns can be set to any scalar type owing to Java's {@link #toString()} method performing * the conversion, as long as the string representation fits into the size constraints (if any) for the string * column.
** Additionally, scalar columns can take single-element array arguments, just like * {@link #setElement(int, int, Object)}. *
* * @param row the zero-based row index * @param col the zero-based column index * @param o the new value to set. For array columns this must match the Java array type * exactly, but for scalar columns additional flexibility is provided for fuzzy * type matching (see description above). * * @throws FitsException if the column could not be set * @throws IllegalArgumentException if the argument cannot be converted to a value for the specified column type. * * @since 1.18 * * @see #get(int, int) */ public void set(int row, int col, Object o) throws FitsException, IllegalArgumentException { ColumnDesc c = columns.get(col); if (o == null) { // Only logicals and strings support 'null' values if (!c.isSingleton()) { throw new TableException("No null values allowed for column of " + c.getLegacyBase() + " arrays."); } else if (c.isString()) { setElement(row, col, ""); } else { setLogical(row, col, null); } } else if (o.getClass().isArray()) { Class> eType = ArrayFuncs.getBaseClass(o); if (!c.getFitsBase().isAssignableFrom(eType) && c.isNumeric()) { o = ArrayFuncs.convertArray(o, c.getFitsBase(), c.getQuantizer()); } setElement(row, col, o); } else if (o instanceof String) { setString(row, col, (String) o); } else if (!c.isSingleton()) { throw new TableException("Cannot set scalar values in non-scalar columns"); } else if (c.isString()) { setElement(row, col, o.toString()); } else if (o instanceof Boolean) { setLogical(row, col, (Boolean) o); } else if (o instanceof Character) { setCharacter(row, col, (Character) o); } else if (o instanceof Number) { setNumber(row, col, (Number) o); } else if (o instanceof ComplexValue) { setElement(row, col, o); } else { throw new IllegalArgumentException("Unsupported scalar type: " + o.getClass()); } } /** * Sets a scalar table entry to the specified numerical value. * * @param row the zero-based row index * @param col the zero-based column index * @param value the new number value * * @throws ClassCastException if the specified column in not a numerical scalar type. * @throws FitsException if the table element could not be altered * * @see #getNumber(int, int) * @see #set(int, int, Object) * * @since 1.18 */ private void setNumber(int row, int col, Number value) throws FitsException, ClassCastException { ColumnDesc c = columns.get(col); // Already checked before calling... // if (!c.isSingleton()) { // throw new ClassCastException("Cannot set scalar value for array column " + col); // } if (c.isLogical()) { Boolean b = null; if (!Double.isNaN(value.doubleValue())) { b = value.longValue() != 0; } setTableElement(row, col, new byte[] {FitsEncoder.byteForBoolean(b)}); return; } Class> base = c.getLegacyBase(); // quantize / unquantize as necessary... Quantizer q = c.getQuantizer(); if (q != null) { boolean decimalBase = (base == float.class || base == double.class); boolean decimalValue = (value instanceof Float || value instanceof Double || value instanceof BigInteger || value instanceof BigDecimal); if (decimalValue && !decimalBase) { value = q.toLong(value.doubleValue()); } else if (!decimalValue && decimalBase) { value = q.toDouble(value.longValue()); } } Object wrapped = null; if (base == byte.class) { wrapped = new byte[] {value.byteValue()}; } else if (base == short.class) { wrapped = new short[] {value.shortValue()}; } else if (base == int.class) { wrapped = new int[] {value.intValue()}; } else if (base == long.class) { wrapped = new long[] {value.longValue()}; } else if (base == float.class) { wrapped = new float[] {value.floatValue()}; } else if (base == double.class) { wrapped = new double[] {value.doubleValue()}; } else { // This could be a char based column... throw new ClassCastException("Cannot set number value for column of type " + base); } setTableElement(row, col, wrapped); } /** * Sets a boolean scalar table entry to the specified value. * * @param row the zero-based row index * @param col the zero-based column index * @param value the new boolean value * * @throws ClassCastException if the specified column in not a boolean scalar type. * @throws FitsException if the table element could not be altered * * @see #getLogical(int, int) * @see #set(int, int, Object) * * @since 1.18 */ private void setLogical(int row, int col, Boolean value) throws FitsException, ClassCastException { ColumnDesc c = columns.get(col); // Already checked before calling... // if (!c.isSingleton()) { // throw new ClassCastException("Cannot set scalar value for array column " + col); // } if (c.isLogical()) { setTableElement(row, col, new byte[] {FitsEncoder.byteForBoolean(value)}); } else if (c.getLegacyBase() == char.class) { setTableElement(row, col, new char[] {value == null ? '\0' : (value ? 'T' : 'F')}); } else { setNumber(row, col, value == null ? Double.NaN : (value ? 1 : 0)); } } /** * Sets a Unicode character scalar table entry to the specified value. * * @param row the zero-based row index * @param col the zero-based column index * @param value the new Unicode character value * * @throws ClassCastException if the specified column in not a boolean scalar type. * @throws FitsException if the table element could not be altered * * @see #getString(int, int) * * @since 1.18 */ private void setCharacter(int row, int col, Character value) throws FitsException, ClassCastException { ColumnDesc c = columns.get(col); // Already checked before calling... // if (!c.isSingleton()) { // throw new IllegalArgumentException("Cannot set scalar value for array column " + col); // } if (c.isLogical()) { setLogical(row, col, FitsUtil.parseLogical(value.toString())); } else if (c.fitsBase == char.class) { setTableElement(row, col, new char[] {value}); } else if (c.fitsBase == byte.class) { setTableElement(row, col, new byte[] {(byte) (value & FitsIO.BYTE_MASK)}); } else { throw new ClassCastException("Cannot convert char value to " + c.fitsBase.getName()); } } /** * Sets a table entry to the specified string value. Scalar column will attempt to parse the value, while *byte[]
andchar[] byte
columns will parse the string as a number (not as a single ASCII character). * * @param row the zero-based row index * @param col the zero-based column index * @param value the new boolean value * * @throws ClassCastException if the specified column is not a scalar type, and neither it is a *byte[]
orchar[] column. * @throws IllegalArgumentException if the String is too long to contain in the column. * @throws NumberFormatException if the numerical value could not be parsed. * @throws FitsException if the table element could not be altered * * @see #getString(int, int) * @see #set(int, int, Object) * * @since 1.18 */ private void setString(int row, int col, String value) throws FitsException, ClassCastException, IllegalArgumentException, NumberFormatException { ColumnDesc c = columns.get(col); // Already checked before calling... // if (!c.isSingleton()) { // throw new IllegalArgumentException("Cannot set scalar value for array column " + col); // } if (c.isLogical()) { setLogical(row, col, FitsUtil.parseLogical(value)); } else if (value.length() == 1) { setCharacter(row, col, value.charAt(0)); } else if (c.fitsDimension() > 1) { throw new ClassCastException("Cannot convert String to multi-dimensional array"); } else if (c.fitsDimension() == 1) { if (c.fitsBase != char.class && c.fitsBase != byte.class) { throw new ClassCastException("Cannot cast String to " + c.fitsBase.getName()); } int len = c.isVariableSize() ? value.length() : c.fitsCount; if (value.length() > len) { throw new IllegalArgumentException("String size " + value.length() + " exceeds entry size of " + len); } if (c.fitsBase == char.class) { setTableElement(row, col, Arrays.copyOf(value.toCharArray(), len)); } else { setTableElement(row, col, FitsUtil.stringToByteArray(value, len)); } } else { try { setNumber(row, col, Long.parseLong(value)); } catch (NumberFormatException e) { setNumber(row, col, Double.parseDouble(value)); } } } /** * @deprecated (for internal use) It may be reduced to private visibility in the future. Sets a * column with the data already flattened. * * @param col The index of the column to be replaced. * @param data The new data array. This should be a one-d primitive array. * * @throws FitsException Thrown if the type of length of the replacement data differs from the original. */ public void setFlattenedColumn(int col, Object data) throws FitsException { ensureData(); Object oldCol = table.getColumn(col); if (data.getClass() != oldCol.getClass() || Array.getLength(data) != Array.getLength(oldCol)) { throw new TableException("Replacement column mismatch at column:" + col); } table.setColumn(col, javaToFits1D(columns.get(col), data)); } @Override public void setRow(int row, Object[] data) throws FitsException { ensureData(); if (data.length != getNCols()) { throw new TableException("Mismatched number of columns: " + data.length + ", expected " + getNCols()); } for (int col = 0; col < data.length; col++) { set(row, col, data[col]); } } /** * @deprecated It is not entirely foolproof for keeping the header in sync -- it is better to (re)wrap tables in a * new HDU after column deletions, and then edit the new header as necessary to incorporate custom * entries. May be removed from the API in the future. */ @Override public void updateAfterDelete(int oldNcol, Header hdr) throws FitsException { hdr.addValue(Standard.NAXIS1, rowLen); int l = 0; for (ColumnDesc d : columns) { d.offset = l; l += d.rowLen(); } } @SuppressWarnings("resource") @Override public void write(ArrayDataOutput os) throws FitsException { try { if (isDeferred() && os == getRandomAccessInput()) { // It it's a deferred mode re-write, then data were edited in place if at all, // so we can skip the main table. ((RandomAccess) os).skipAllBytes(getRegularTableSize()); } else { // otherwise make sure we loaded all data before writing to the output ensureData(); // Write the regular table (if any) if (getRegularTableSize() > 0) { table.write(os); } } // Now check if we need to write the heap if (getParameterSize() > 0) { for (long rem = getHeapOffset(); rem > 0;) { byte[] b = new byte[(int) Math.min(getHeapOffset(), 1 << Short.SIZE)]; os.write(b); rem -= b.length; } getHeap().write(os); if (heapReserve > 0) { byte[] b = new byte[heapReserve]; os.write(b); } } FitsUtil.pad(os, getTrueSize(), (byte) 0); } catch (IOException e) { throw new FitsException("Unable to write table:" + e, e); } } /** * Returns the heap offset component from a pointer. * * @param p the pointer, either a int[2]
or along[2]
. * * @return the offset component from the pointer */ private long getPointerOffset(Object p) { return (p instanceof long[]) ? ((long[]) p)[1] : ((int[]) p)[1]; } /** * Returns the number of elements reported in a heap pointer. * * @param p the pointer, either aint[2]
or along[2]
. * * @return the element count component from the pointer */ private long getPointerCount(Object p) { return (p instanceof long[]) ? ((long[]) p)[0] : ((int[]) p)[0]; } /** * Puts a FITS data array onto our heap, returning its locator pointer. The data will overwrite the previous heap * entry, if provided, so long as the new data fits in the same place. Otherwise the new data is placed at the end * of the heap. * * @param c The column descriptor, specifying the data type * @param o The variable-length data * @param p The heap pointer, where this element was stored on the heap before, ornull
if * we aren't replacing an earlier entry. * * @return the heap pointer information, eitherint[2]
or else along[2]
* * @throws FitsException if the data could not be accessed in full from the heap. */ @SuppressFBWarnings(value = "RR_NOT_CHECKED", justification = "not propagated or used locally") private Object putOnHeap(ColumnDesc c, Object o, Object oldPointer) throws FitsException { return putOnHeap(getHeap(), c, o, oldPointer); } /** * Puts a FITS data array onto a specific heap, returning its locator pointer. The data will overwrite the previous * heap entry, if provided, so long as the new data fits in the same place. Otherwise the new data is placed at the * end of the heap. * * @param h The heap object to use. * @param c The column descriptor, specifying the data type * @param o The variable-length data in Java form. * @param p The heap pointer, where this element was stored on the heap before, ornull
if * we aren't replacing an earlier entry. * * @return the heap pointer information, eitherint[2]
or else along[2]
* * @throws FitsException if the data could not be accessed in full from the heap. */ @SuppressFBWarnings(value = "RR_NOT_CHECKED", justification = "not propagated or used locally") private Object putOnHeap(FitsHeap h, ColumnDesc c, Object o, Object oldPointer) throws FitsException { // Flatten data for heap o = ArrayFuncs.flatten(o); // By default put data at the end of the heap; int off = h.size(); // The number of Java elements is the same as the number of FITS elements, except for strings and complex // numbers int len = (c.isComplex() || c.isString()) ? -1 : Array.getLength(o); // Convert to FITS storage array o = javaToFits1D(c, o); // For complex values and strings, determine length from converted object.... if (len < 0) { len = Array.getLength(o); // If complex in primitive 1D form, then length is half the number of elements. if (c.isComplex() && o.getClass().getComponentType().isPrimitive()) { len >>>= 1; } } if (oldPointer != null) { if (len <= getPointerCount(oldPointer)) { // Write data back at the old heap location off = (int) getPointerOffset(oldPointer); } } h.putData(o, off); return c.hasLongPointers() ? new long[] {len, off} : new int[] {len, off}; } /** * Returns a FITS data array from the heap * * @param c The column descriptor, specifying the data type * @param p The heap pointer, eitherint[2]
or else along[2]
* @param isEnhanced Whether logicals should be returned as {@link Boolean} (rather thanboolean
) * and complex values as {@link ComplexValue} (rather thanfloat[2]
or *double[2]
), or arrays thereof. Methods prior to 1.18 should set this to *false
for back compatible behavior. * * @return the FITS array object retrieved from the heap * * @throws FitsException if the data could not be accessed in full from the heap. */ protected Object getFromHeap(ColumnDesc c, Object p, boolean isEnhanced) throws FitsException { long len = getPointerCount(p); long off = getPointerOffset(p); if (off > Integer.MAX_VALUE || len > Integer.MAX_VALUE) { throw new FitsException("Data located beyond 32-bit accessible heap limit: off=" + off + ", len=" + len); } Object e = null; if (c.isComplex()) { e = Array.newInstance(c.getFitsBase(), (int) len, 2); } else { e = Array.newInstance(c.getFitsBase(), c.getFitsBaseCount((int) len)); } readHeap(off, e); return fitsToJava1D(c, e, (int) len, isEnhanced); } /** * Convert Java arrays to their FITS representation. Transformation include boolean → 'T'/'F' or '\0'; * Strings → byte arrays; variable length arrays → pointers (after writing data to heap). * * @param c The column descritor * @param o A one-dimensional Java array * * @return An one-dimensional array with values as stored in FITS. * * @throws FitsException if the operation failed */ private Object javaToFits1D(ColumnDesc c, Object o) throws FitsException { if (c.isBits()) { return FitsUtil.bitsToBytes((boolean[]) o); } if (c.isLogical()) { // Convert true/false to 'T'/'F', or null to '\0' return FitsUtil.booleansToBytes(o); } if (c.isComplex()) { if (o instanceof ComplexValue || o instanceof ComplexValue[]) { return ArrayFuncs.complexToDecimals(o, c.fitsBase); } } if (c.isString()) { // Convert strings to array of bytes. if (o == null) { if (c.isVariableSize()) { return new byte[0]; } return Array.newInstance(byte.class, c.fitsShape); } if (o instanceof String) { int l = c.getStringLength(); if (l < 0) { // Not fixed width, write the whole string. l = ((String) o).length(); } return FitsUtil.stringToByteArray((String) o, l); } if (c.isVariableSize() && c.delimiter != 0) { // Write variable-length string arrays in delimited form for (String s : (String[]) o) { // We set the string length to that of the longest element + 1 c.setStringLength(Math.max(c.stringLength, s == null ? 1 : s.length() + 1)); } return FitsUtil.stringsToDelimitedBytes((String[]) o, c.getStringLength(), c.delimiter); } // Fixed length substring array (not delimited). // For compatibility with tools that do not process array dimension, ASCII NULL should not // be used between components (permissible only at the end of all strings) return FitsUtil.stringsToByteArray((String[]) o, c.getStringLength(), FitsUtil.BLANK_SPACE); } return boxedToArray(o); } /** * Converts from the FITS representation of data to their basic Java array representation. * * @param c The column descritor * @param o A one-dimensional array of values as stored in FITS * @param bits A bit count for bit arrays (otherwise unused). * @param isEnhanced Whether logicals should be returned as {@link Boolean} (rather thanboolean
) * and complex values as {@link ComplexValue} (rather thanfloat[2]
or *double[2]
), or arrays thereof. Methods prior to 1.18 should set this to *false
for back compatible behavior. * * @return A {@link String} or a one-dimensional array with the matched basic Java type * * @throws FitsException if the operation failed */ private Object fitsToJava1D(ColumnDesc c, Object o, int bits, boolean isEnhanced) { if (c.isBits()) { return FitsUtil.bytesToBits((byte[]) o, bits); } if (c.isLogical()) { return isEnhanced ? FitsUtil.bytesToBooleanObjects(o) : FitsUtil.byteToBoolean((byte[]) o); } if (c.isComplex() && isEnhanced) { return ArrayFuncs.decimalsToComplex(o); } if (c.isString()) { byte[] bytes = (byte[]) o; int len = c.getStringLength(); if (c.isVariableSize()) { if (c.delimiter != 0) { // delimited array of strings return FitsUtil.delimitedBytesToStrings(bytes, c.getStringLength(), c.delimiter); } } // If fixed or variable length arrays of strings... if (c.isSingleton()) { // Single fixed string -- get it all but trim trailing spaces return FitsUtil.extractString(bytes, new ParsePosition(0), bytes.length, FitsUtil.ASCII_NULL); } // Array of fixed-length strings -- we trim trailing spaces in each component String[] s = new String[bytes.length / len]; for (int i = 0; i < s.length; i++) { s[i] = FitsUtil.extractString(bytes, new ParsePosition(i * len), len, FitsUtil.ASCII_NULL); } return s; } return o; } /** * Create a column table with the specified number of rows. This is used when we defer instantiation of the * ColumnTable until the user requests data from the table. * * @param rows the number of rows to allocate * * @throws FitsException if the operation failed */ protected void createTable(int rows) throws FitsException { int nfields = columns.size(); Object[] data = new Object[nfields]; int[] sizes = new int[nfields]; for (int i = 0; i < nfields; i++) { ColumnDesc c = columns.get(i); sizes[i] = c.getTableBaseCount(); data[i] = c.newInstance(rows); } table = createColumnTable(data, sizes); nRow = rows; } private static Object boxedToArray(Object o) throws FitsException { if (o.getClass().isArray()) { return o; } // Convert boxed types to primitive arrays of 1. if (o instanceof Number) { if (o instanceof Byte) { return new byte[] {(byte) o}; } if (o instanceof Short) { return new short[] {(short) o}; } if (o instanceof Integer) { return new int[] {(int) o}; } if (o instanceof Long) { return new long[] {(long) o}; } if (o instanceof Float) { return new float[] {(float) o}; } if (o instanceof Double) { return new double[] {(double) o}; } throw new FitsException("Unsupported Number type: " + o.getClass()); } if (o instanceof Boolean) { return new Boolean[] {(Boolean) o}; } if (o instanceof Character) { return new char[] {(char) o}; } return o; } /** * Sets the input to use for reading (and possibly writing) this table. If the input implements * {@link ReadWriteAccess}, then it can be used for both reading and (re)writing the data, including editing in * deferred mode. * * @param in The input from which we can read the table data. */ private void setInput(ArrayDataInput in) { encoder = (in instanceof ReadWriteAccess) ? new FitsEncoder((ReadWriteAccess) in) : null; } @Override public void read(ArrayDataInput in) throws FitsException { setInput(in); super.read(in); } @Override protected void loadData(ArrayDataInput in) throws IOException, FitsException { setInput(in); createTable(nRow); readTrueData(in); } /** * Extracts a column descriptor from the FITS header for a given column index * * @param header the FITS header containing the column description(s) * @param col zero-based column index * * @return the Descriptor for that column. * * @throws FitsException if the header deswcription is invalid or incomplete */ public static ColumnDesc getDescriptor(Header header, int col) throws FitsException { String tform = header.getStringValue(Standard.TFORMn.n(col + 1)); if (tform == null) { throw new FitsException("Missing TFORM" + (col + 1)); } int count = 1; char type = 0; ParsePosition pos = new ParsePosition(0); try { count = AsciiFuncs.parseInteger(tform, pos); } catch (Exception e) { // Keep going... } try { type = Character.toUpperCase(AsciiFuncs.extractChar(tform, pos)); } catch (Exception e) { throw new FitsException("Missing data type in TFORM: [" + tform + "]"); } ColumnDesc c = new ColumnDesc(); if (header.containsKey(Standard.TTYPEn.n(col + 1))) { c.name(header.getStringValue(Standard.TTYPEn.n(col + 1))); } if (type == POINTER_INT || type == POINTER_LONG) { // Variable length column... c.setVariableSize(type == POINTER_LONG); // Get the data type... try { type = Character.toUpperCase(AsciiFuncs.extractChar(tform, pos)); } catch (Exception e) { throw new FitsException("Missing variable-length data type in TFORM: [" + tform + "]"); } } // The special types... if (type == 'C' || type == 'M') { c.isComplex = true; } else if (type == 'X') { c.isBits = true; } if (!c.setFitsType(type)) { throw new FitsException("Invalid type '" + type + "' in column:" + col); } if (!c.isVariableSize()) { // Fixed sized column... int[] dims = parseTDims(header.getStringValue(Standard.TDIMn.n(col + 1))); if (dims == null) { c.setFitsShape((count == 1 && type != 'A') ? SINGLETON_SHAPE : new int[] {count}); c.stringLength = -1; // T.B.D. further below... } else { c.setFitsShape(dims); } } if (c.isString()) { // For vairable-length columns or of TDIM was not defined determine substring length from TFORM. c.parseSubstringConvention(tform, pos, c.getStringLength() < 0); } // Force to use the count in the header, even if it does not match up with the dimension otherwise. c.fitsCount = count; c.quant = Quantizer.fromTableHeader(header, col); if (c.quant.isDefault()) { c.quant = null; } return c; } /** * Process one column from a FITS Header. * * @throws FitsException if the operation failed */ private int processCol(Header header, int col, int offset) throws FitsException { ColumnDesc c = getDescriptor(header, col); c.offset = offset; columns.add(c); return c.rowLen(); } /** * @deprecated (for internal use) Used Only by {@link nom.tam.image.compression.hdu.CompressedTableData} so * it would make a better private method in there.. ` */ protected void addByteVaryingColumn() { addColumn(ColumnDesc.createForVariableSize(byte.class)); } /** * @deprecated (for internal use) This method should have visibility reduced to private */ @SuppressWarnings("javadoc") protected ColumnTable> createColumnTable(Object[] arrCol, int[] sizes) throws TableException { return new ColumnTable<>(arrCol, sizes); } /** * Returns the heap, after initializing it from the input as necessary * * @return the initialized heap * * @throws FitsException if we had trouble initializing it from the input. */ @SuppressWarnings("resource") private synchronized FitsHeap getHeap() throws FitsException { if (heap == null) { readHeap(getRandomAccessInput()); } return heap; } /** * Reads an array from the heap. Subclasses may override this, for example to provide read-only access to a related * table's heap area. * * @param offset the heap offset * @param array the array to populate from the heap area * * @throws FitsException if there was an issue accessing the heap */ protected void readHeap(long offset, Object array) throws FitsException { getHeap().getData((int) offset, array); } /** * Read the heap which contains the data for variable length arrays. A. Kovacs (4/1/08) Separated heap reading, s.t. * the heap can be properly initialized even if in deferred read mode. columnToArray() checks and initializes the * heap as necessary. * * @param input stream to read from. * * @throws FitsException if the heap could not be read from the stream * * @deprecated (for internal use) unused. */ protected synchronized void readHeap(ArrayDataInput input) throws FitsException { if (input instanceof RandomAccess) { FitsUtil.reposition(input, getFileOffset() + getHeapAddress()); } heap = new FitsHeap(heapFileSize); if (input != null) { heap.read(input); } } /** * Read table, heap and padding * * @param i the stream to read the data from. * * @throws FitsException if the reading failed */ protected synchronized void readTrueData(ArrayDataInput i) throws FitsException { try { table.read(i); i.skipAllBytes(getHeapOffset()); if (heap == null) { readHeap(i); } } catch (IOException e) { throw new FitsException("Error reading binary table data:" + e, e); } } /** * Check if the column number is valid. * * @param j The Java index (first=0) of the column to check. * * @returntrue
if the column is valid */ protected boolean validColumn(int j) { return j >= 0 && j < getNCols(); } /** * Check to see if this is a valid row. * * @param i The Java index (first=0) of the row to check. * * @returntrue
if the row is valid */ protected boolean validRow(int i) { return getNRows() > 0 && i >= 0 && i < getNRows(); } /** * @deprecated (for internal use) Visibility should be reduced to protected. */ @Override public void fillHeader(Header h) throws FitsException { fillHeader(h, true); } /** * Fills (updates) the essential header description of this table in the header, optionally updating the essential * column descriptions also if desired. * * @param h The FITS header to populate * @param updateColumns Whether to update the essential column descriptions also * * @throws FitsException if there was an error accessing the header. */ void fillHeader(Header h, boolean updateColumns) throws FitsException { h.deleteKey(Standard.SIMPLE); h.deleteKey(Standard.EXTEND); Standard.context(BinaryTable.class); Cursorc = h.iterator(); c.add(HeaderCard.create(Standard.XTENSION, Standard.XTENSION_BINTABLE)); c.add(HeaderCard.create(Standard.BITPIX, Bitpix.BYTE.getHeaderValue())); c.add(HeaderCard.create(Standard.NAXIS, 2)); c.add(HeaderCard.create(Standard.NAXIS1, rowLen)); c.add(HeaderCard.create(Standard.NAXIS2, nRow)); if (h.getLongValue(Standard.PCOUNT, -1L) < getParameterSize()) { c.add(HeaderCard.create(Standard.PCOUNT, getParameterSize())); } c.add(HeaderCard.create(Standard.GCOUNT, 1)); c.add(HeaderCard.create(Standard.TFIELDS, columns.size())); if (getHeapOffset() == 0) { h.deleteKey(Standard.THEAP); } else { c.add(HeaderCard.create(Standard.THEAP, getHeapAddress())); } if (updateColumns) { for (int i = 0; i < columns.size(); i++) { c.setKey(Standard.TFORMn.n(i + 1).key()); fillForColumn(h, c, i); } } Standard.context(null); } /** * Update the header to reflect the details of a given column. * * @throws FitsException if the operation failed */ void fillForColumn(Header header, Cursor hc, int col) throws FitsException { ColumnDesc c = columns.get(col); try { Standard.context(BinaryTable.class); if (c.name() != null) { hc.add(HeaderCard.create(Standard.TTYPEn.n(col + 1), c.name())); } hc.add(HeaderCard.create(Standard.TFORMn.n(col + 1), c.getTFORM())); String tdim = c.getTDIM(); if (tdim != null) { hc.add(HeaderCard.create(Standard.TDIMn.n(col + 1), tdim)); } if (c.quant != null) { c.quant.editTableHeader(header, col); } } finally { Standard.context(null); } } /** * Returns the column descriptor of a given column in this table * * @param column the zero-based column index * * @return the column's descriptor * * @throws ArrayIndexOutOfBoundsException if this table does not contain a column with that index. * * @see #getDescriptor(String) */ public ColumnDesc getDescriptor(int column) throws ArrayIndexOutOfBoundsException { return columns.get(column); } /** * Returns the (first) column descriptor whose name matches the specified value. * * @param name The column name (case sensitive). * * @return The descriptor of the first column by that name, or null
if the table contains no * column by that name. * * @see #getDescriptor(int) * @see #indexOf(String) * * @since 1.20 * * @author Attila Kovacs */ public ColumnDesc getDescriptor(String name) { int col = indexOf(name); return col < 0 ? null : getDescriptor(col); } /** * Converts a column from FITS logical values to bits. Null values (allowed in logical columns) will map to *false
. * * @param col The zero-based index of the column to be reset. * * @return Whether the conversion was possible. * * * @since 1.18 */ public boolean convertToBits(int col) { ColumnDesc c = columns.get(col); if (c.isBits) { return true; } if (c.base != boolean.class) { return false; } c.isBits = true; return true; } /** * Convert a column from float/double to float complex/double complex. This is only possible for certain columns. * The return status indicates if the conversion is possible. * * @param index The zero-based index of the column to be reset. * * @return Whether the conversion is possible. * * * @throws FitsException if the operation failed * * @since 1.18 * * @see ColumnDesc#isComplex() * @see #addComplexColumn(Object, Class) */ public boolean setComplexColumn(int index) throws FitsException { if (!validColumn(index)) { return false; } ColumnDesc c = columns.get(index); if (c.isComplex()) { return true; } if (c.base != float.class && c.base != double.class) { return false; } if (!c.isVariableSize()) { if (c.getLastFitsDim() != 2) { return false; } // Set the column to complex c.isComplex = true; // Update the legacy (wrapped array) shape c.setLegacyShape(c.fitsShape); return true; } // We need to make sure that for every row, there are // an even number of elements so that we can // convert to an integral number of complex numbers. for (int i = 1; i < nRow; i++) { if (getPointerCount(getRawElement(i, index)) % 2 != 0) { return false; } } // Halve the length component of array descriptors (2 reals = 1 complex) for (int i = 1; i < nRow; i++) { Object p = getRawElement(i, index); long len = getPointerCount(p) >>> 1; if (c.hasLongPointers()) { ((long[]) p)[0] = len; } else { ((int[]) p)[0] = (int) len; } setTableElement(i, index, p); } // Set the column to complex c.isComplex = true; return true; } /** * Checks if this table contains a heap for storing variable length arrays (VLAs). * * @returntrue
if the table contains a heap, or elsefalse
. * * @since 1.19.1 */ public final boolean containsHeap() { return getParameterSize() > 0; } /** ** Defragments the heap area of this table, compacting the heap area, and returning the number of bytes by which the * heap size has been reduced. When tables with variable-sized columns are modified, the heap may retain old data as * columns are removed or elements get replaced with new data of different size. The data order in the heap may also * get jumbled, causing what would appear to be sequential reads to jump all over the heap space with the caching. * And, depending on how the heap was constructed in the first place, it may not be optimal for the row-after-row * table access that is the most typical use case. *
** This method rebuilds the heap by taking elements in table read order (by rows, and columns) and puts them on a * new heap. *
** For best squential read performance, you should defragment all tables that have been built column-by-column * before writing them to a FITS file. The only time defragmentation is really not needed is if the table was built * row-by-row, with no modifications to variable-length content after the fact. *
* * @return the number of bytes by which the heap has shrunk as a result of defragmentation. * * @throws FitsException if there was an error accessing the heap or the main data table comntaining the heap * locators. In case of an error the table content may be left in a damaged state. * * @see #compact() * @see #setElement(int, int, Object) * @see #addColumn(Object) * @see #deleteColumns(int, int) * @see #setColumn(int, Object) * * @since 1.18 */ public synchronized long defragment() throws FitsException { if (!containsHeap()) { return 0L; } int[] eSize = new int[columns.size()]; for (int j = 0; j < columns.size(); j++) { ColumnDesc c = columns.get(j); if (c.isVariableSize()) { eSize[j] = ElementType.forClass(c.getFitsBase()).size(); } } FitsHeap hp = getHeap(); long oldSize = hp.size(); FitsHeap compact = new FitsHeap(0); for (int i = 0; i < nRow; i++) { for (int j = 0; j < columns.size(); j++) { ColumnDesc c = columns.get(j); if (c.isVariableSize()) { Object p = getRawElement(i, j); int len = (int) getPointerCount(p); // Copy to new heap... int pos = compact.copyFrom(hp, (int) getPointerOffset(p), c.getFitsBaseCount(len) * eSize[j]); // Same length as before... if (p instanceof long[]) { ((long[]) p)[1] = pos; } else { ((int[]) p)[1] = pos; } // Update pointers in table setTableElement(i, j, p); } } } heap = compact; return oldSize - compact.size(); } /** * Discard the information about the original heap size (if this table was read from an input), and instead use the * real size of the actual heap (plus reserved space around it) when writing to an output. Compacted tables may not * be re-writeable to the same file from which they were read, since they may be shorter than the original, but they * can always be written to a different file, which may at times be smaller than the original. It may be used along * with {@link #defragment()} to create FITS files with optimized storage from FITS files that may contain wasted * space. * * @see #defragment() * * @since 1.19.1 * * @author Attila Kovacs */ public synchronized void compact() { heapFileSize = 0; } @Override public BinaryTableHDU toHDU() throws FitsException { Header h = new Header(); fillHeader(h); return new BinaryTableHDU(h, this); } }