com.squarespace.template.expr.Expr Maven / Gradle / Ivy
Go to download
Show more of this group Show more artifacts with this name
Show all versions of template-core Show documentation
Show all versions of template-core Show documentation
Squarespace template compiler
package com.squarespace.template.expr;
import static com.squarespace.template.expr.Conversions.asbool;
import static com.squarespace.template.expr.Conversions.asint;
import static com.squarespace.template.expr.Conversions.asliteral;
import static com.squarespace.template.expr.Conversions.asnode;
import static com.squarespace.template.expr.Conversions.asnum;
import static com.squarespace.template.expr.Conversions.asstr;
import static com.squarespace.template.expr.Conversions.ch;
import static com.squarespace.template.expr.Conversions.decimali;
import static com.squarespace.template.expr.Conversions.hexi;
import static com.squarespace.template.expr.Operations.mul;
import static com.squarespace.template.expr.Tokens.MINUS_ONE;
import static com.squarespace.template.expr.Tokens.bool;
import static com.squarespace.template.expr.Tokens.num;
import static com.squarespace.template.expr.Tokens.str;
import java.util.ArrayList;
import java.util.Collections;
import java.util.HashMap;
import java.util.List;
import java.util.Map;
import org.apache.commons.lang3.StringEscapeUtils;
import com.fasterxml.jackson.databind.JsonNode;
import com.fasterxml.jackson.databind.node.BooleanNode;
import com.fasterxml.jackson.databind.node.DoubleNode;
import com.fasterxml.jackson.databind.node.NullNode;
import com.fasterxml.jackson.databind.node.TextNode;
import com.squarespace.template.Context;
import com.squarespace.template.ErrorInfo;
import com.squarespace.template.ExecuteErrorType;
import com.squarespace.template.GeneralUtils;
import com.squarespace.template.TokenMatcher;
/**
* Expression evaluation using an extended version of Dijkstra's "shunting
* yard" algorithm with JavaScript semantics. This algorithm was chosen as
* it is simple, sufficient, and can be implemented compactly, minimizing
* the size of the code and making it easier to verify correct.
*
* Features:
*
* - Number, Boolean, Null, and String types
* - String parsing of backslash char, hex, and unicode escapes
* - unicode escapes can contain up to 6 hex characters with a
* value of <= 0x10FFFF
* - "\n", "\x20", "\u2018", "\u01f600" are all valid strings
* - Decimal and hex numbers
* - Assignment statements
* - Variable references
* - Operator precedence
* - Common math, binary and logical operators
* - Both strict and non-strict equality
* - Nesting of sub-expressions with parenthesis
* - Single pass evaluation of multiple semicolon-delimited expressions
* - String equality and concatenation
* - Function calls with a variable number of arguments
* - example: max(1, 2, 3, ..., N) is accepted
* - Type conversion functions
* - Common constants (pi, e, etc)
* - Configurable parse and reduce limits
* - maximum tokens an expression can contain
* - maximum length of a concatenated string
*
* The algorithm assumes the expression is well-formed and performs minimal
* validation. A malformed expression should either terminate evaluation
* early or produce no value.
*/
public class Expr {
private static final String E_MISMATCHED_OP = "Mismatched operator found:";
private static final String E_UNEXPECTED_OPERATOR = "Unexpected operator found during evaluation:";
private static final ExprOptions DEFAULT_OPTIONS = new ExprOptions();
// Value for the PI constant in the expression engine
public static final double PI = 3.141592653589793;
// Value for the E constant in the expression engine
public static final double E = 2.718281828459045;
// Expressions ready to evaluate
private final List> expr = new ArrayList<>();
// Errors that occur during tokenization or assembly. Evaluation errors will
// be appended directly to the context's errors array.
private final List errors = new ArrayList<>();
// Temporary storage to build expressions
private final Stack tokens = new Stack<>();
// reasonable limits to bound the computation
private final int maxTokens;
private final int maxStringLen;
private final TokenMatcher matcher;
private final String raw;
/**
* Construct an expression from the given string. Construction will
* only tokenize the input.
*/
public Expr(String raw) {
this(raw, null);
}
/**
* Construct an expression from the given string and options.
* Construction will only tokenize the input.
*/
public Expr(String raw, ExprOptions options) {
if (options == null) {
options = DEFAULT_OPTIONS;
}
this.maxTokens = options.maxTokens();
this.maxStringLen = options.maxStringLen();
this.matcher = new TokenMatcher(raw);
this.raw = raw;
this.tokenize(raw, 0, raw.length());
}
@Override
public String toString() {
return "Expression[" + new TextNode(this.raw) + "]";
}
/**
* Errors that occur when parsing and assembling the expression.
* Runtime errors are appended directly to the Context.
*/
public List errors() {
return this.errors;
}
/**
* Tokens parsed from the input string.
*/
public Stack tokens() {
return this.tokens;
}
/**
* Expressions assembled from the input tokens.
*/
public List> expressions() {
return this.expr;
}
/**
* Built-in functions.
*TODO: future, make these configurable
*/
private final Map FUNCTIONS = new HashMap() {
{
put("max", Functions::max);
put("min", Functions::min);
put("abs", Functions::abs);
put("num", Functions::num);
put("str", Functions::str);
put("bool", Functions::bool);
}
};
/**
* Built-in constants.
* TODO: future, make these configurable
*/
private final Map CONSTANTS = new HashMap(){{
put("null", Tokens.NULL);
put("true", Tokens.TRUE);
put("false", Tokens.FALSE);
put("PI", Tokens.PI);
put("E", Tokens.E);
put("Infinity", Tokens.INFINITY);
put("NaN", Tokens.NAN);
}};
/**
* Reduce each of the expressions to its simplest form, then return the final value
* of the last expression as output.
*/
public JsonNode reduce(Context ctx) {
int len = this.expr.size();
JsonNode r = null;
for (int i = 0; i < len; i++) {
List expr = this.expr.get(i);
r = this.reduceExpr(ctx, expr);
}
return r;
}
/**
* Reduce an expression to its simplest form.
*/
public JsonNode reduceExpr(Context ctx, List expr) {
Stack stack = new Stack<>();
loop: for (Token t : expr) {
switch (t.type) {
case BOOLEAN:
case STRING:
case NUMBER:
case NULL:
case VARIABLE:
case ARGS:
stack.push(t);
continue;
case CALL: {
List args = new ArrayList<>();
Token top = stack.top();
// Pop values from the stack until we hit the argument delimiter.
while (top != null && top.type != ExprTokenType.ARGS) {
Token arg = asliteral(ctx, stack.pop());
if (arg != null) {
args.add(arg);
}
top = stack.top();
}
// Pop the argument delimiter
stack.pop();
// Reverse the arguments and apply them.
Collections.reverse(args);
// Get a reference to the function implementation.
CallToken call = (CallToken)t;
FunctionDef fimpl = FUNCTIONS.get(call.name);
// The function is guaranteed to exist here, since its existence
// was verified when the call token was constructed.
Token r = fimpl.apply(args);
if (r == null) {
ErrorInfo error = ctx.error(ExecuteErrorType.EXPRESSION_REDUCE)
.data("Error calling function " + call.name);
ctx.addError(error);
break loop;
}
// Push the result onto the stack.
stack.push(r);
continue;
}
case OPERATOR: {
// Unary operators directly manipulate top of stack to avoid a pop-then-push
Operator o = ((OperatorToken)t).value;
switch (o.type) {
case MINUS: {
Token arg = asliteral(ctx, stack.top());
stack.top(arg == null ? null : mul(MINUS_ONE, arg));
continue;
}
case PLUS: {
// unary plus casts the argument to number but doesn't change sign
Token arg = asliteral(ctx, stack.top());
stack.top(arg == null ? null : num(asnum(arg)));
continue;
}
case LNOT: {
Token arg = asliteral(ctx, stack.top());
stack.top(arg == null ? null : bool(!asbool(arg)));
continue;
}
case BNOT: {
Token arg = asliteral(ctx, stack.top());
stack.top(arg == null ? null : num(~asint(arg)));
continue;
}
case ASN: {
Token b = asliteral(ctx, stack.pop());
Token a = stack.pop();
// Make sure the arguments to the assignment are valid
if (a != null && a.type == ExprTokenType.VARIABLE && b != null) {
Object[] name = ((VarToken)a).name;
if (name != null && name.length == 1 && name[0] instanceof String && ((String)name[0]).charAt(0) == '@') {
// Set the variable in the context.
ctx.setVar((String)name[0], asnode(b));
}
}
// When an assignment operator is encountered, we consider the expression
// complete. This leaves no result.
stack.top(null);
break loop;
}
default:
// fall through to handle all binary operators
}
// Binary operators, pop 2 args from the stack and push result
Token b = asliteral(ctx, stack.pop());
Token a = asliteral(ctx, stack.pop());
// Validate operator args are present and valid
if (a == null || b == null) {
// Invalid arguments to operator, bail out.
ErrorInfo error = ctx.error(ExecuteErrorType.EXPRESSION_REDUCE)
.data("Invalid arguments to operator " + o.desc);
ctx.addError(error);
break loop;
}
Token r = null;
switch (o.type) {
case MUL:
r = mul(a, b);
break;
case DIV: {
double v = asnum(b);
r = num(v == 0 ? Double.NaN : asnum(a) / v);
break;
}
case ADD:
// Numeric addition or string concatenation.
if (a.type == ExprTokenType.STRING || b.type == ExprTokenType.STRING) {
// Convert both arguments to string
String _a = asstr(a);
String _b = asstr(b);
// Ensure a concatenated string won't exceed the configured limit.
if (this.maxStringLen > 0 && (_a.length() + _b.length() > this.maxStringLen)) {
ErrorInfo error = ctx.error(ExecuteErrorType.EXPRESSION_REDUCE)
.data("Concatenation would exceed maximum string length " + this.maxStringLen);
ctx.addError(error);
break loop;
}
r = str(_a + _b);
} else {
r = num(asnum(a) + asnum(b));
}
break;
case SUB:
r = num(asnum(a) - asnum(b));
break;
case POW:
r = num(Math.pow(asnum(a), asnum(b)));
break;
case MOD: {
double v = asnum(b);
r = num(v == 0 ? Double.NaN : asnum(a) % v);
break;
}
case SHL:
r = num(asint(a) << asint(b));
break;
case SHR:
r = num(asint(a) >> asint(b));
break;
case LT:
r = Operations.lt(a, b);
break;
case LTEQ:
r = Operations.lteq(a, b);
break;
case GT:
r = Operations.gt(a, b);
break;
case GTEQ:
r = Operations.gteq(a, b);
break;
case EQ:
r = Operations.eq(a, b);
break;
case NEQ:
r = Operations.neq(a, b);
break;
case SEQ:
r = Operations.seq(a, b);
break;
case SNEQ:
r = Operations.sneq(a, b);
break;
case BAND:
r = num(asint(a) & asint(b));
break;
case BXOR:
r = num(asint(a) ^ asint(b));
break;
case BOR:
r = num(asint(a) | asint(b));
break;
case LAND:
r = bool(asbool(a) && asbool(b));
break;
case LOR:
r = bool(asbool(a) || asbool(b));
break;
default:
// all unary and binary operators should be handled above.
// other operators (parenthesis, semicolon) are eliminated when
// the expressions are assembled.
this.errors.add(E_UNEXPECTED_OPERATOR + " " + o.desc);
stack.top(null);
break loop;
}
// Push result onto stack
stack.push(r);
}
}
}
// Return a valid literal from the top of the stack, or null
// if an unexpected token is present.
Token r = stack.top();
if (r != null) {
// Ensure the value is a literal
Token v = asliteral(ctx, r);
if (v != null) {
// We have a supported value.
switch (v.type) {
case BOOLEAN:
return ((BooleanToken)v).value ? BooleanNode.TRUE : BooleanNode.FALSE;
case NUMBER:
return new DoubleNode(((NumberToken)v).value);
case STRING:
return new TextNode(((StringToken)v).value);
case NULL:
return NullNode.getInstance();
default:
// Fall through
break;
}
}
// The token was an unexpected type, which is an error
ErrorInfo error = ctx.error(ExecuteErrorType.EXPRESSION_REDUCE)
.data("Reduce error: unexpected token on stack");
ctx.addError(error);
}
return null;
}
/**
* Iterate over tokens and build expressions using the shunting yard algorithm.
*/
public void build() {
// If any tokenization errors occurred, refuse to assemble the expression. The
// token array is likely incomplete and therefore the expression should be
// considered invalid.
if (!this.errors.isEmpty()) {
return;
}
// Output stack containing final RPN expression
List out = new ArrayList<>();
// Stack used to shunt operator tokens during expression construction
Stack ops = new Stack<>();
for (Token t : this.tokens.elems()) {
switch (t.type) {
case OPERATOR: {
Operator o = ((OperatorToken)t).value;
switch (o.type) {
case SEMI:
// Semicolons separate multiple expressions. Push the expression
// we've accumulated and reset the state.
this.pushExpr(out, ops);
out = new ArrayList<>();
break;
case LPRN:
// Opens a nested expression or function call
ops.push(t);
break;
case COMMA: {
// Argument separator outputs all non-operators until we hit
// a left parenthesis
Token top = ops.top();
while (cond1(top)) {
out.add(ops.pop());
top = ops.top();
}
break;
}
case RPRN: {
// Output all non-operator tokens until we hit the matching
// left parenthesis.
Token top = ops.top();
while (cond1(top)) {
out.add(ops.pop());
top = ops.top();
}
// Ensure parenthesis are balanced.
if (top == null ||
top.type != ExprTokenType.OPERATOR ||
((OperatorToken)top).value.type != OperatorType.LPRN) {
this.errors.add(E_MISMATCHED_OP + " " + ((OperatorToken)t).value.desc);
return;
}
ops.pop();
// If a function call token preceeded the left parenthesis, pop it to the output
top = ops.top();
if (top != null && top.type == ExprTokenType.CALL) {
out.add(ops.pop());
}
break;
}
default: {
// We have an operator. Before we can send it to the output
// we need to pop all other operators with higher precedence,
// or the same precedence with left associativity. We also stop
// at non-operators and left parenthesis.
Token top = ops.top();
while (cond2(top, o.prec)) {
out.add(ops.pop());
top = ops.top();
}
ops.push(t);
break;
}
}
break;
}
case CALL:
// Delimit the end of the argument list for the function call
out.add(Tokens.ARGS);
// Push the call onto the operator stack. Once all arguments have
// been output we pop the call and output it.
ops.push(t);
break;
default:
out.add(t);
break;
}
}
this.pushExpr(out, ops);
}
/**
* Condition for build() method.
*/
private boolean cond1(Token t) {
if (t == null) {
return false;
}
if (t.type != ExprTokenType.OPERATOR) {
return false;
}
Operator o = ((OperatorToken)t).value;
return o.type != OperatorType.LPRN;
}
/**
* Condition for build() method.
*/
private boolean cond2(Token t, int prec) {
if (t == null) {
return false;
}
if (t.type != ExprTokenType.OPERATOR) {
return false;
}
Operator o = ((OperatorToken)t).value;
return o.type != OperatorType.LPRN && (o.prec > prec ||
(o.prec == prec && o.assoc == Assoc.LEFT));
}
/**
* Push a token.
*/
private void push(Token t) {
// If an error occurs, we stop accepting tokens.
if (!this.errors.isEmpty()) {
return;
}
if (t.type == ExprTokenType.OPERATOR) {
OperatorToken ot = (OperatorToken)t;
switch (ot.value.type) {
// convert unary plus / minus
case SUB:
case ADD: {
Token top = this.tokens.top();
if (top == null ||
(top.type == ExprTokenType.OPERATOR &&
((OperatorToken)top).value.type != OperatorType.RPRN)) {
t = ot.value.type == OperatorType.SUB ? Operators.MINUS : Operators.PLUS;
}
break;
}
case LPRN: {
Token top = this.tokens.top();
if (top != null &&
top.type == ExprTokenType.VARIABLE &&
((VarToken)top).name != null &&
((VarToken)top).name.length == 1) {
// Check if name corresponds to a valid built-in function, and
// convert the name to a function call token.
Object name = ((VarToken)top).name[0];
if (FUNCTIONS.containsKey(name)) {
this.tokens.top(new CallToken((String)name));
} else {
this.errors.add("Invalid function: " + name);
return;
}
}
break;
}
default:
break;
}
}
this.tokens.push(t);
if (this.maxTokens > 0 && this.tokens.length() > this.maxTokens) {
this.errors.add("Expression exceeds the maximum number of allowed tokens: " + this.maxTokens);
}
}
/**
* Push an expression.
*/
private void pushExpr(List queue, Stack ops) {
while (ops.length() > 0) {
Token t = ops.pop();
// We detect unexpected operators here.
if (t.type == ExprTokenType.OPERATOR) {
OperatorToken o = (OperatorToken)t;
switch (o.value.type) {
case LPRN:
case RPRN:
this.errors.add(E_MISMATCHED_OP + " " + o.value.desc);
return;
default:
break;
}
}
queue.add(t);
}
if (!queue.isEmpty()) {
this.expr.add(queue);
}
}
/**
* Tokenize the string input.
*/
public void tokenize(String str, int i, int len) {
loop: while (i < len) {
char c0 = str.charAt(i);
char c1 = ch(str, i + 1);
switch (c0) {
case '0':
case '1':
case '2':
case '3':
case '4':
case '5':
case '6':
case '7':
case '8':
case '9':
// check for a hexadecimal number prefix
boolean hx = c0 == '0' && (c1 == 'x' || c1 == 'X');
i = hx ? this.hex(str, i + 2, len) : this.decimal(str, i, len);
if (i < 0) {
// the hex / decimal parse methods emit errors, so no need to
break loop;
}
continue;
case '"':
case '\'':
i = string(str, i + 1, len, c0);
if (i == -1) {
break loop;
}
continue;
case '*':
if (c1 == '*') {
i++;
this.push(Operators.POW);
} else {
this.push(Operators.MUL);
}
break;
case '/':
this.push(Operators.DIV);
break;
case '%':
this.push(Operators.MOD);
break;
case '+':
this.push(Operators.ADD);
break;
case '-':
this.push(Operators.SUB);
break;
case '=':
if (c1 == '=') {
i++;
if (ch(str, i + 1) == '=') {
i++;
this.push(Operators.SEQ);
} else {
this.push(Operators.EQ);
}
} else {
this.push(Operators.ASN);
}
break;
case '!':
if (c1 == '=') {
i++;
if (ch(str, i + 1) == '=') {
i++;
this.push(Operators.SNEQ);
} else {
this.push(Operators.NEQ);
}
} else {
this.push(Operators.LNOT);
}
break;
case '<':
if (c1 == '<') {
i++;
this.push(Operators.SHL);
} else if (c1 == '=') {
i++;
this.push(Operators.LTEQ);
} else {
this.push(Operators.LT);
}
break;
case '>':
if (c1 == '>') {
i++;
this.push(Operators.SHR);
} else if (c1 == '=') {
i++;
this.push(Operators.GTEQ);
} else {
this.push(Operators.GT);
}
break;
case '~':
this.push(Operators.BNOT);
break;
case '&':
if (c1 == '&') {
i++;
this.push(Operators.LAND);
} else {
this.push(Operators.BAND);
}
break;
case '|':
if (c1 == '|') {
i++;
this.push(Operators.LOR);
} else {
this.push(Operators.BOR);
}
break;
case '^':
this.push(Operators.BXOR);
break;
case ' ':
case '\n':
case '\t':
case '\r':
case '\u00a0':
break;
case ',':
this.push(Operators.COMMA);
break;
case ';':
this.push(Operators.SEMI);
break;
case '(':
this.push(Operators.LPRN);
break;
case ')':
this.push(Operators.RPRN);
break;
default: {
// Here we handle variable references, and named functions / constants.
// We use the matchVariable to cover these cases and disambiguate them.
// Function names are whitelisted and must be immediately followed by an
// open parenthesis.
matcher.region(i, len);
if (matcher.variable()) {
int end = matcher.matchEnd();
Object[] raw = GeneralUtils.splitVariable(str.substring(i, end));
i = end;
if (raw != null && raw.length == 1 && raw[0] instanceof String) {
// Names for constants. These names can conflict with references to
// context variables on the immediate node, e.g. { "PI": "apple" }.
// To disambiguate, use references of the form "@.PI" or bind
// local variables before calling the expression.
String name = (String)raw[0];
Token value = CONSTANTS.get(name);
if (value != null) {
this.push(value);
continue;
}
// Fall through and assume this is a variable reference. If followed
// immediately by a left parenthesis it may be a function call, but
// we determine that in the push() method.
}
this.push(new VarToken(raw));
continue;
}
// input character we can't handle
this.errors.add("Unexpected " + charName(c0) + " at " + i + ": " + StringEscapeUtils.escapeJava("" + c0));
i = -1;
break;
}
}
if (i == -1) {
// Error occurred, bail out
break;
}
i++;
}
}
/**
* Scan a decimal number and push a token, or an error message.
*/
private int decimal(String str, int i, int len) {
int j = decimali(str, i, len);
switch (j) {
case -2:
this.errors.add("Expected a digit after exponent in decimal number");
break;
case -3:
this.errors.add("Duplicate decimal point in number");
break;
case -4:
this.errors.add("Unexpected decimal point in exponent");
break;
}
if (j < 0) {
return -1;
}
// no need to consider radix as numbers will always be in decimal
String text = str.substring(i, j);
this.push(new NumberToken(Double.parseDouble(text)));
return j;
}
/**
* Parse a hexadecimal integer number.
*/
private int hex(String str, int i, int len) {
int j = hexi(str, i, len);
if (i == j) {
this.errors.add("Expected digits after start of hex number");
return -1;
}
String text = str.substring(i, j);
double value = Conversions.hexnum(text);
this.push(new NumberToken(value));
return j;
}
/**
* Parse a string literal.
*/
private int string(String str, int i, int len, char end) {
// Accumulate decoded characters
StringBuilder s = new StringBuilder();
int j;
while (i < len) {
char c = str.charAt(i);
j = i + 1;
// Handle backslash escape sequences
if (c == '\\' && j < len) {
c = str.charAt(j);
switch (c) {
// Newline
case 'n':
s.append('\n');
break;
// Tab
case 't':
s.append('\t');
break;
// Form feed
case 'f':
s.append('\f');
break;
// Carriage return
case 'r':
s.append('\r');
break;
// Byte
case 'x': {
// Skip over escape
i += 2;
// Check if we have enough characters to parse the full escape
int lim = i + 2;
if (lim >= len) {
this.errors.add(E_INVALID_HEX);
return -1;
}
// find end of hex char sequence
int k = hexi(str, i, lim);
if (k != lim) {
this.errors.add(E_INVALID_HEX);
return -1;
}
// Decode range of chars as 2-digit hex number
int code = Integer.parseInt(str.substring(i, k), 16);
// Eliminate unwanted ascii control bytes here.
if (code <= 0x08 || (code >= 0x0e && code < 0x20)) {
// emit replacement char
s.append(' ');
} else {
// if k === lim above, hex string is valid
s.append((char)code);
}
// Skip over escape sequence and continue
i = k;
continue;
}
// Unicode character escape 4 or 8 digits '\u0000' or '\U00000000'
case 'u':
case 'U': {
// Skip over escape
i += 2;
// a unicode escape can contain 4 or 8 characters.
int lim = i + (c == 'u' ? 4 : 8);
// find end of hex char sequence
int k = hexi(str, i, lim < len ? lim : len);
// escape sequence end must match limit
if (k != lim) {
this.errors.add(E_INVALID_UNICODE);
return -1;
}
// Decode range of chars as 4- or 8-digit hex number. It is possible
// for an 8-digit hex value to exceed the range of int, so we parse
// as long and then constrain with a conditional.
String repr = str.substring(i, k);
long code = Long.parseLong(repr, 16);
// Eliminate unwanted ascii control byte here. Also eliminate
// out of range invalid Unicode characters.
if (code <= 0x08 || (code >= 0x0e && code < 0x20) || code > 0x10FFFF) {
// emit replacement char
s.append(' ');
} else if (code > 0xFFFF) {
// convert to a surrogate pair
code -= 0x10000;
s.append((char)(((code / 0x400) | 0) + 0xd800));
s.append((char)((code % 0x400) + 0xdc00));
} else {
// append the char directly
s.append((char)code);
}
// Skip over escape sequence and continue
i = k;
continue;
}
// Literal character
default:
s.append(c);
break;
}
// If we're here, the escape was length 2, so skip it
i += 2;
continue;
}
// If we've found the matching string delimiter, push the string token
if (c == end) {
i++;
this.push(new StringToken(s.toString()));
return i;
}
// Bare line separators aren't allowed in strings
switch (c) {
case '\n':
case '\r':
this.errors.add("Illegal bare " + charName(c) + " character in string literal");
return -1;
}
// append the character to the output and advance
s.append(c);
i++;
}
// Matching end delimiter was never found
this.errors.add("Unterminated string");
return -1;
}
private static final Map CHARS = new HashMap(){{
put('\b', "backspace");
put('\f', "form feed");
put('\n', "line feed");
put('\r', "carriage return");
put('\t', "tab");
}};
private static final String charName(char c) {
String r = CHARS.get(c);
return r != null ? r : c <= 0x001f ? "control character" : "character";
}
// Error messages used more than once
private static final String E_INVALID_HEX = "Invalid 2-char hex escape found";
private static final String E_INVALID_UNICODE = "Invalid unicode escape found";
}
© 2015 - 2024 Weber Informatics LLC | Privacy Policy