org.antlr.v4.runtime.CodePointCharStream Maven / Gradle / Ivy
Go to download
Show more of this group Show more artifacts with this name
Show all versions of virtdata-lib-curves4 Show documentation
Show all versions of virtdata-lib-curves4 Show documentation
Statistical sampling library for use in virtdata libraries, based
on apache commons math 4
/*
* Copyright (c) 2012-2017 The ANTLR Project. All rights reserved.
* Use of this file is governed by the BSD 3-clause license that
* can be found in the LICENSE.txt file in the project root.
*/
package org.antlr.v4.runtime;
import org.antlr.v4.runtime.misc.Interval;
import java.nio.charset.StandardCharsets;
/**
* Alternative to {@link ANTLRInputStream} which treats the input
* as a series of Unicode code points, instead of a series of UTF-16
* code units.
*
* Use this if you need to parse input which potentially contains
* Unicode values > U+FFFF.
*/
public abstract class CodePointCharStream implements CharStream {
protected final int size;
protected final String name;
// To avoid lots of virtual method calls, we directly access
// the state of the underlying code points in the
// CodePointBuffer.
protected int position;
// Use the factory method {@link #fromBuffer(CodePointBuffer)} to
// construct instances of this type.
private CodePointCharStream(int position, int remaining, String name) {
// TODO
assert position == 0;
this.size = remaining;
this.name = name;
this.position = 0;
}
// Visible for testing.
abstract Object getInternalStorage();
/**
* Constructs a {@link CodePointCharStream} which provides access
* to the Unicode code points stored in {@code codePointBuffer}.
*/
public static CodePointCharStream fromBuffer(CodePointBuffer codePointBuffer) {
return fromBuffer(codePointBuffer, UNKNOWN_SOURCE_NAME);
}
/**
* Constructs a named {@link CodePointCharStream} which provides access
* to the Unicode code points stored in {@code codePointBuffer}.
*/
public static CodePointCharStream fromBuffer(CodePointBuffer codePointBuffer, String name) {
// Java lacks generics on primitive types.
//
// To avoid lots of calls to virtual methods in the
// very hot codepath of LA() below, we construct one
// of three concrete subclasses.
//
// The concrete subclasses directly access the code
// points stored in the underlying array (byte[],
// char[], or int[]), so we can avoid lots of virtual
// method calls to ByteBuffer.get(offset).
switch (codePointBuffer.getType()) {
case BYTE:
return new CodePoint8BitCharStream(
codePointBuffer.position(),
codePointBuffer.remaining(),
name,
codePointBuffer.byteArray(),
codePointBuffer.arrayOffset());
case CHAR:
return new CodePoint16BitCharStream(
codePointBuffer.position(),
codePointBuffer.remaining(),
name,
codePointBuffer.charArray(),
codePointBuffer.arrayOffset());
case INT:
return new CodePoint32BitCharStream(
codePointBuffer.position(),
codePointBuffer.remaining(),
name,
codePointBuffer.intArray(),
codePointBuffer.arrayOffset());
}
throw new UnsupportedOperationException("Not reached");
}
@Override
public final void consume() {
if (size - position == 0) {
assert LA(1) == IntStream.EOF;
throw new IllegalStateException("cannot consume EOF");
}
position = position + 1;
}
@Override
public final int index() {
return position;
}
@Override
public final int size() {
return size;
}
/** mark/release do nothing; we have entire buffer */
@Override
public final int mark() {
return -1;
}
@Override
public final void release(int marker) {
}
@Override
public final void seek(int index) {
position = index;
}
@Override
public final String getSourceName() {
if (name == null || name.isEmpty()) {
return UNKNOWN_SOURCE_NAME;
}
return name;
}
@Override
public final String toString() {
return getText(Interval.of(0, size - 1));
}
// 8-bit storage for code points <= U+00FF.
private static final class CodePoint8BitCharStream extends CodePointCharStream {
private final byte[] byteArray;
private CodePoint8BitCharStream(int position, int remaining, String name, byte[] byteArray, int arrayOffset) {
super(position, remaining, name);
// TODO
assert arrayOffset == 0;
this.byteArray = byteArray;
}
/** Return the UTF-16 encoded string for the given interval */
@Override
public String getText(Interval interval) {
int startIdx = Math.min(interval.a, size);
int len = Math.min(interval.b - interval.a + 1, size - startIdx);
// We know the maximum code point in byteArray is U+00FF,
// so we can treat this as if it were ISO-8859-1, aka Latin-1,
// which shares the same code points up to 0xFF.
return new String(byteArray, startIdx, len, StandardCharsets.ISO_8859_1);
}
@Override
public int LA(int i) {
int offset;
switch (Integer.signum(i)) {
case -1:
offset = position + i;
if (offset < 0) {
return IntStream.EOF;
}
return byteArray[offset] & 0xFF;
case 0:
// Undefined
return 0;
case 1:
offset = position + i - 1;
if (offset >= size) {
return IntStream.EOF;
}
return byteArray[offset] & 0xFF;
}
throw new UnsupportedOperationException("Not reached");
}
@Override
Object getInternalStorage() {
return byteArray;
}
}
// 16-bit internal storage for code points between U+0100 and U+FFFF.
private static final class CodePoint16BitCharStream extends CodePointCharStream {
private final char[] charArray;
private CodePoint16BitCharStream(int position, int remaining, String name, char[] charArray, int arrayOffset) {
super(position, remaining, name);
this.charArray = charArray;
// TODO
assert arrayOffset == 0;
}
/** Return the UTF-16 encoded string for the given interval */
@Override
public String getText(Interval interval) {
int startIdx = Math.min(interval.a, size);
int len = Math.min(interval.b - interval.a + 1, size - startIdx);
// We know there are no surrogates in this
// array, since otherwise we would be given a
// 32-bit int[] array.
//
// So, it's safe to treat this as if it were
// UTF-16.
return new String(charArray, startIdx, len);
}
@Override
public int LA(int i) {
int offset;
switch (Integer.signum(i)) {
case -1:
offset = position + i;
if (offset < 0) {
return IntStream.EOF;
}
return charArray[offset] & 0xFFFF;
case 0:
// Undefined
return 0;
case 1:
offset = position + i - 1;
if (offset >= size) {
return IntStream.EOF;
}
return charArray[offset] & 0xFFFF;
}
throw new UnsupportedOperationException("Not reached");
}
@Override
Object getInternalStorage() {
return charArray;
}
}
// 32-bit internal storage for code points between U+10000 and U+10FFFF.
private static final class CodePoint32BitCharStream extends CodePointCharStream {
private final int[] intArray;
private CodePoint32BitCharStream(int position, int remaining, String name, int[] intArray, int arrayOffset) {
super(position, remaining, name);
this.intArray = intArray;
// TODO
assert arrayOffset == 0;
}
/** Return the UTF-16 encoded string for the given interval */
@Override
public String getText(Interval interval) {
int startIdx = Math.min(interval.a, size);
int len = Math.min(interval.b - interval.a + 1, size - startIdx);
// Note that we pass the int[] code points to the String constructor --
// this is supported, and the constructor will convert to UTF-16 internally.
return new String(intArray, startIdx, len);
}
@Override
public int LA(int i) {
int offset;
switch (Integer.signum(i)) {
case -1:
offset = position + i;
if (offset < 0) {
return IntStream.EOF;
}
return intArray[offset];
case 0:
// Undefined
return 0;
case 1:
offset = position + i - 1;
if (offset >= size) {
return IntStream.EOF;
}
return intArray[offset];
}
throw new UnsupportedOperationException("Not reached");
}
@Override
Object getInternalStorage() {
return intArray;
}
}
}