package com.fasterxml.jackson.core.json;
import java.io.*;
import com.fasterxml.jackson.core.*;
import com.fasterxml.jackson.core.base.ParserBase;
import com.fasterxml.jackson.core.io.CharTypes;
import com.fasterxml.jackson.core.io.IOContext;
import com.fasterxml.jackson.core.sym.ByteQuadsCanonicalizer;
import com.fasterxml.jackson.core.util.*;
import static com.fasterxml.jackson.core.JsonTokenId.*;
This is a concrete implementation of JsonParser
, which is based on a InputStream
as the input source. /**
* This is a concrete implementation of {@link JsonParser}, which is
* based on a {@link java.io.InputStream} as the input source.
*/
public class UTF8StreamJsonParser
extends ParserBase
{
final static byte BYTE_LF = (byte) '\n';
@SuppressWarnings("deprecation")
private final static int FEAT_MASK_TRAILING_COMMA = Feature.ALLOW_TRAILING_COMMA.getMask();
@SuppressWarnings("deprecation")
private final static int FEAT_MASK_LEADING_ZEROS = Feature.ALLOW_NUMERIC_LEADING_ZEROS.getMask();
@SuppressWarnings("deprecation")
private final static int FEAT_MASK_NON_NUM_NUMBERS = Feature.ALLOW_NON_NUMERIC_NUMBERS.getMask();
@SuppressWarnings("deprecation")
private final static int FEAT_MASK_ALLOW_MISSING = Feature.ALLOW_MISSING_VALUES.getMask();
private final static int FEAT_MASK_ALLOW_SINGLE_QUOTES = Feature.ALLOW_SINGLE_QUOTES.getMask();
private final static int FEAT_MASK_ALLOW_UNQUOTED_NAMES = Feature.ALLOW_UNQUOTED_FIELD_NAMES.getMask();
private final static int FEAT_MASK_ALLOW_JAVA_COMMENTS = Feature.ALLOW_COMMENTS.getMask();
private final static int FEAT_MASK_ALLOW_YAML_COMMENTS = Feature.ALLOW_YAML_COMMENTS.getMask();
// This is the main input-code lookup table, fetched eagerly
private final static int[] _icUTF8 = CharTypes.getInputCodeUtf8();
// Latin1 encoding is not supported, but we do use 8-bit subset for
// pre-processing task, to simplify first pass, keep it fast.
protected final static int[] _icLatin1 = CharTypes.getInputCodeLatin1();
/*
/**********************************************************
/* Configuration
/**********************************************************
*/
Codec used for data binding when (if) requested; typically full
ObjectMapper
, but that abstract is not part of core
package.
/**
* Codec used for data binding when (if) requested; typically full
* <code>ObjectMapper</code>, but that abstract is not part of core
* package.
*/
protected ObjectCodec _objectCodec;
Symbol table that contains field names encountered so far
/**
* Symbol table that contains field names encountered so far
*/
final protected ByteQuadsCanonicalizer _symbols;
/*
/**********************************************************
/* Parsing state
/**********************************************************
*/
Temporary buffer used for name parsing.
/**
* Temporary buffer used for name parsing.
*/
protected int[] _quadBuffer = new int[16];
Flag that indicates that the current token has not yet
been fully processed, and needs to be finished for
some access (or skipped to obtain the next token)
/**
* Flag that indicates that the current token has not yet
* been fully processed, and needs to be finished for
* some access (or skipped to obtain the next token)
*/
protected boolean _tokenIncomplete;
Temporary storage for partially parsed name bytes.
/**
* Temporary storage for partially parsed name bytes.
*/
private int _quad1;
Value of ParserBase._inputPtr
at the time when the first character of name token was read. Used for calculating token location when requested; combined with ParserBase._currInputProcessed
, may be updated appropriately as needed. Since: 2.7
/**
* Value of {@link #_inputPtr} at the time when the first character of
* name token was read. Used for calculating token location when requested;
* combined with {@link #_currInputProcessed}, may be updated appropriately
* as needed.
*
* @since 2.7
*/
protected int _nameStartOffset;
Since: 2.7
/**
* @since 2.7
*/
protected int _nameStartRow;
Since: 2.7
/**
* @since 2.7
*/
protected int _nameStartCol;
/*
/**********************************************************
/* Input buffering (from former 'StreamBasedParserBase')
/**********************************************************
*/
protected InputStream _inputStream;
/*
/**********************************************************
/* Current input data
/**********************************************************
*/
Current buffer from which data is read; generally data is read into
buffer from input source, but in some cases pre-loaded buffer
is handed to the parser.
/**
* Current buffer from which data is read; generally data is read into
* buffer from input source, but in some cases pre-loaded buffer
* is handed to the parser.
*/
protected byte[] _inputBuffer;
Flag that indicates whether the input buffer is recycable (and
needs to be returned to recycler once we are done) or not.
If it is not, it also means that parser can NOT modify underlying
buffer.
/**
* Flag that indicates whether the input buffer is recycable (and
* needs to be returned to recycler once we are done) or not.
*<p>
* If it is not, it also means that parser can NOT modify underlying
* buffer.
*/
protected boolean _bufferRecyclable;
/*
/**********************************************************
/* Life-cycle
/**********************************************************
*/
Deprecated: Since 2.10
/**
* @deprecated Since 2.10
*/
@Deprecated
public UTF8StreamJsonParser(IOContext ctxt, int features, InputStream in,
ObjectCodec codec, ByteQuadsCanonicalizer sym,
byte[] inputBuffer, int start, int end,
boolean bufferRecyclable)
{
this(ctxt, features, in, codec, sym,
inputBuffer, start, end, 0, bufferRecyclable);
}
public UTF8StreamJsonParser(IOContext ctxt, int features, InputStream in,
ObjectCodec codec, ByteQuadsCanonicalizer sym,
byte[] inputBuffer, int start, int end, int bytesPreProcessed,
boolean bufferRecyclable)
{
super(ctxt, features);
_inputStream = in;
_objectCodec = codec;
_symbols = sym;
_inputBuffer = inputBuffer;
_inputPtr = start;
_inputEnd = end;
_currInputRowStart = start - bytesPreProcessed;
// If we have offset, need to omit that from byte offset, so:
_currInputProcessed = -start + bytesPreProcessed;
_bufferRecyclable = bufferRecyclable;
}
@Override
public ObjectCodec getCodec() {
return _objectCodec;
}
@Override
public void setCodec(ObjectCodec c) {
_objectCodec = c;
}
/*
/**********************************************************
/* Overrides for life-cycle
/**********************************************************
*/
@Override
public int releaseBuffered(OutputStream out) throws IOException
{
int count = _inputEnd - _inputPtr;
if (count < 1) {
return 0;
}
// let's just advance ptr to end
int origPtr = _inputPtr;
_inputPtr += count;
out.write(_inputBuffer, origPtr, count);
return count;
}
@Override
public Object getInputSource() {
return _inputStream;
}
/*
/**********************************************************
/* Overrides, low-level reading
/**********************************************************
*/
protected final boolean _loadMore() throws IOException
{
if (_inputStream != null) {
int space = _inputBuffer.length;
if (space == 0) { // only occurs when we've been closed
return false;
}
int count = _inputStream.read(_inputBuffer, 0, space);
if (count > 0) {
final int bufSize = _inputEnd;
_currInputProcessed += bufSize;
_currInputRowStart -= bufSize;
// 26-Nov-2015, tatu: Since name-offset requires it too, must offset
// this increase to avoid "moving" name-offset, resulting most likely
// in negative value, which is fine as combine value remains unchanged.
_nameStartOffset -= bufSize;
_inputPtr = 0;
_inputEnd = count;
return true;
}
// End of input
_closeInput();
// Should never return 0, so let's fail
if (count == 0) {
throw new IOException("InputStream.read() returned 0 characters when trying to read "+_inputBuffer.length+" bytes");
}
}
return false;
}
@Override
protected void _closeInput() throws IOException
{
// We are not to call close() on the underlying InputStream
// unless we "own" it, or auto-closing feature is enabled.
if (_inputStream != null) {
if (_ioContext.isResourceManaged() || isEnabled(Feature.AUTO_CLOSE_SOURCE)) {
_inputStream.close();
}
_inputStream = null;
}
}
Method called to release internal buffers owned by the base reader. This may be called along with _closeInput
(for example, when explicitly closing this reader instance), or separately (if need be). /**
* Method called to release internal buffers owned by the base
* reader. This may be called along with {@link #_closeInput} (for
* example, when explicitly closing this reader instance), or
* separately (if need be).
*/
@Override
protected void _releaseBuffers() throws IOException
{
super._releaseBuffers();
// Merge found symbols, if any:
_symbols.release();
if (_bufferRecyclable) {
byte[] buf = _inputBuffer;
if (buf != null) {
// Let's not set it to null; this way should get slightly more meaningful
// error messages in case someone closes parser indirectly, without realizing.
if (buf != NO_BYTES) {
_inputBuffer = NO_BYTES;
_ioContext.releaseReadIOBuffer(buf);
}
}
}
}
/*
/**********************************************************
/* Public API, data access
/**********************************************************
*/
@Override
public String getText() throws IOException
{
if (_currToken == JsonToken.VALUE_STRING) {
if (_tokenIncomplete) {
_tokenIncomplete = false;
return _finishAndReturnString(); // only strings can be incomplete
}
return _textBuffer.contentsAsString();
}
return _getText2(_currToken);
}
@Override // since 2.8
public int getText(Writer writer) throws IOException
{
JsonToken t = _currToken;
if (t == JsonToken.VALUE_STRING) {
if (_tokenIncomplete) {
_tokenIncomplete = false;
_finishString(); // only strings can be incomplete
}
return _textBuffer.contentsToWriter(writer);
}
if (t == JsonToken.FIELD_NAME) {
String n = _parsingContext.getCurrentName();
writer.write(n);
return n.length();
}
if (t != null) {
if (t.isNumeric()) {
return _textBuffer.contentsToWriter(writer);
}
char[] ch = t.asCharArray();
writer.write(ch);
return ch.length;
}
return 0;
}
// // // Let's override default impls for improved performance
// @since 2.1
@Override
public String getValueAsString() throws IOException
{
if (_currToken == JsonToken.VALUE_STRING) {
if (_tokenIncomplete) {
_tokenIncomplete = false;
return _finishAndReturnString(); // only strings can be incomplete
}
return _textBuffer.contentsAsString();
}
if (_currToken == JsonToken.FIELD_NAME) {
return getCurrentName();
}
return super.getValueAsString(null);
}
// @since 2.1
@Override
public String getValueAsString(String defValue) throws IOException
{
if (_currToken == JsonToken.VALUE_STRING) {
if (_tokenIncomplete) {
_tokenIncomplete = false;
return _finishAndReturnString(); // only strings can be incomplete
}
return _textBuffer.contentsAsString();
}
if (_currToken == JsonToken.FIELD_NAME) {
return getCurrentName();
}
return super.getValueAsString(defValue);
}
// since 2.6
@Override
public int getValueAsInt() throws IOException
{
JsonToken t = _currToken;
if ((t == JsonToken.VALUE_NUMBER_INT) || (t == JsonToken.VALUE_NUMBER_FLOAT)) {
// inlined 'getIntValue()'
if ((_numTypesValid & NR_INT) == 0) {
if (_numTypesValid == NR_UNKNOWN) {
return _parseIntValue();
}
if ((_numTypesValid & NR_INT) == 0) {
convertNumberToInt();
}
}
return _numberInt;
}
return super.getValueAsInt(0);
}
// since 2.6
@Override
public int getValueAsInt(int defValue) throws IOException
{
JsonToken t = _currToken;
if ((t == JsonToken.VALUE_NUMBER_INT) || (t == JsonToken.VALUE_NUMBER_FLOAT)) {
// inlined 'getIntValue()'
if ((_numTypesValid & NR_INT) == 0) {
if (_numTypesValid == NR_UNKNOWN) {
return _parseIntValue();
}
if ((_numTypesValid & NR_INT) == 0) {
convertNumberToInt();
}
}
return _numberInt;
}
return super.getValueAsInt(defValue);
}
protected final String _getText2(JsonToken t)
{
if (t == null) {
return null;
}
switch (t.id()) {
case ID_FIELD_NAME:
return _parsingContext.getCurrentName();
case ID_STRING:
// fall through
case ID_NUMBER_INT:
case ID_NUMBER_FLOAT:
return _textBuffer.contentsAsString();
default:
return t.asString();
}
}
@Override
public char[] getTextCharacters() throws IOException
{
if (_currToken != null) { // null only before/after document
switch (_currToken.id()) {
case ID_FIELD_NAME:
if (!_nameCopied) {
String name = _parsingContext.getCurrentName();
int nameLen = name.length();
if (_nameCopyBuffer == null) {
_nameCopyBuffer = _ioContext.allocNameCopyBuffer(nameLen);
} else if (_nameCopyBuffer.length < nameLen) {
_nameCopyBuffer = new char[nameLen];
}
name.getChars(0, nameLen, _nameCopyBuffer, 0);
_nameCopied = true;
}
return _nameCopyBuffer;
case ID_STRING:
if (_tokenIncomplete) {
_tokenIncomplete = false;
_finishString(); // only strings can be incomplete
}
// fall through
case ID_NUMBER_INT:
case ID_NUMBER_FLOAT:
return _textBuffer.getTextBuffer();
default:
return _currToken.asCharArray();
}
}
return null;
}
@Override
public int getTextLength() throws IOException
{
if (_currToken != null) { // null only before/after document
switch (_currToken.id()) {
case ID_FIELD_NAME:
return _parsingContext.getCurrentName().length();
case ID_STRING:
if (_tokenIncomplete) {
_tokenIncomplete = false;
_finishString(); // only strings can be incomplete
}
// fall through
case ID_NUMBER_INT:
case ID_NUMBER_FLOAT:
return _textBuffer.size();
default:
return _currToken.asCharArray().length;
}
}
return 0;
}
@Override
public int getTextOffset() throws IOException
{
// Most have offset of 0, only some may have other values:
if (_currToken != null) {
switch (_currToken.id()) {
case ID_FIELD_NAME:
return 0;
case ID_STRING:
if (_tokenIncomplete) {
_tokenIncomplete = false;
_finishString(); // only strings can be incomplete
}
// fall through
case ID_NUMBER_INT:
case ID_NUMBER_FLOAT:
return _textBuffer.getTextOffset();
default:
}
}
return 0;
}
@Override
public byte[] getBinaryValue(Base64Variant b64variant) throws IOException
{
if (_currToken != JsonToken.VALUE_STRING &&
(_currToken != JsonToken.VALUE_EMBEDDED_OBJECT || _binaryValue == null)) {
_reportError("Current token ("+_currToken+") not VALUE_STRING or VALUE_EMBEDDED_OBJECT, can not access as binary");
}
// To ensure that we won't see inconsistent data, better clear up state...
if (_tokenIncomplete) {
try {
_binaryValue = _decodeBase64(b64variant);
} catch (IllegalArgumentException iae) {
throw _constructError("Failed to decode VALUE_STRING as base64 ("+b64variant+"): "+iae.getMessage());
}
// let's clear incomplete only now; allows for accessing other textual content in error cases
_tokenIncomplete = false;
} else { // may actually require conversion...
if (_binaryValue == null) {
@SuppressWarnings("resource")
ByteArrayBuilder builder = _getByteArrayBuilder();
_decodeBase64(getText(), builder, b64variant);
_binaryValue = builder.toByteArray();
}
}
return _binaryValue;
}
@Override
public int readBinaryValue(Base64Variant b64variant, OutputStream out) throws IOException
{
// if we have already read the token, just use whatever we may have
if (!_tokenIncomplete || _currToken != JsonToken.VALUE_STRING) {
byte[] b = getBinaryValue(b64variant);
out.write(b);
return b.length;
}
// otherwise do "real" incremental parsing...
byte[] buf = _ioContext.allocBase64Buffer();
try {
return _readBinary(b64variant, out, buf);
} finally {
_ioContext.releaseBase64Buffer(buf);
}
}
protected int _readBinary(Base64Variant b64variant, OutputStream out,
byte[] buffer) throws IOException
{
int outputPtr = 0;
final int outputEnd = buffer.length - 3;
int outputCount = 0;
while (true) {
// first, we'll skip preceding white space, if any
int ch;
do {
if (_inputPtr >= _inputEnd) {
_loadMoreGuaranteed();
}
ch = (int) _inputBuffer[_inputPtr++] & 0xFF;
} while (ch <= INT_SPACE);
int bits = b64variant.decodeBase64Char(ch);
if (bits < 0) { // reached the end, fair and square?
if (ch == INT_QUOTE) {
break;
}
bits = _decodeBase64Escape(b64variant, ch, 0);
if (bits < 0) { // white space to skip
continue;
}
}
// enough room? If not, flush
if (outputPtr > outputEnd) {
outputCount += outputPtr;
out.write(buffer, 0, outputPtr);
outputPtr = 0;
}
int decodedData = bits;
// then second base64 char; can't get padding yet, nor ws
if (_inputPtr >= _inputEnd) {
_loadMoreGuaranteed();
}
ch = _inputBuffer[_inputPtr++] & 0xFF;
bits = b64variant.decodeBase64Char(ch);
if (bits < 0) {
bits = _decodeBase64Escape(b64variant, ch, 1);
}
decodedData = (decodedData << 6) | bits;
// third base64 char; can be padding, but not ws
if (_inputPtr >= _inputEnd) {
_loadMoreGuaranteed();
}
ch = _inputBuffer[_inputPtr++] & 0xFF;
bits = b64variant.decodeBase64Char(ch);
// First branch: can get padding (-> 1 byte)
if (bits < 0) {
if (bits != Base64Variant.BASE64_VALUE_PADDING) {
// as per [JACKSON-631], could also just be 'missing' padding
if (ch == INT_QUOTE) {
decodedData >>= 4;
buffer[outputPtr++] = (byte) decodedData;
if (b64variant.usesPadding()) {
--_inputPtr; // to keep parser state bit more consistent
_handleBase64MissingPadding(b64variant);
}
break;
}
bits = _decodeBase64Escape(b64variant, ch, 2);
}
if (bits == Base64Variant.BASE64_VALUE_PADDING) {
// Ok, must get padding
if (_inputPtr >= _inputEnd) {
_loadMoreGuaranteed();
}
ch = _inputBuffer[_inputPtr++] & 0xFF;
if (!b64variant.usesPaddingChar(ch)) {
if (_decodeBase64Escape(b64variant, ch, 3) != Base64Variant.BASE64_VALUE_PADDING) {
throw reportInvalidBase64Char(b64variant, ch, 3, "expected padding character '"+b64variant.getPaddingChar()+"'");
}
}
// Got 12 bits, only need 8, need to shift
decodedData >>= 4;
buffer[outputPtr++] = (byte) decodedData;
continue;
}
}
// Nope, 2 or 3 bytes
decodedData = (decodedData << 6) | bits;
// fourth and last base64 char; can be padding, but not ws
if (_inputPtr >= _inputEnd) {
_loadMoreGuaranteed();
}
ch = _inputBuffer[_inputPtr++] & 0xFF;
bits = b64variant.decodeBase64Char(ch);
if (bits < 0) {
if (bits != Base64Variant.BASE64_VALUE_PADDING) {
// as per [JACKSON-631], could also just be 'missing' padding
if (ch == INT_QUOTE) {
decodedData >>= 2;
buffer[outputPtr++] = (byte) (decodedData >> 8);
buffer[outputPtr++] = (byte) decodedData;
if (b64variant.usesPadding()) {
--_inputPtr; // to keep parser state bit more consistent
_handleBase64MissingPadding(b64variant);
}
break;
}
bits = _decodeBase64Escape(b64variant, ch, 3);
}
if (bits == Base64Variant.BASE64_VALUE_PADDING) {
/* With padding we only get 2 bytes; but we have
* to shift it a bit so it is identical to triplet
* case with partial output.
* 3 chars gives 3x6 == 18 bits, of which 2 are
* dummies, need to discard:
*/
decodedData >>= 2;
buffer[outputPtr++] = (byte) (decodedData >> 8);
buffer[outputPtr++] = (byte) decodedData;
continue;
}
}
// otherwise, our triplet is now complete
decodedData = (decodedData << 6) | bits;
buffer[outputPtr++] = (byte) (decodedData >> 16);
buffer[outputPtr++] = (byte) (decodedData >> 8);
buffer[outputPtr++] = (byte) decodedData;
}
_tokenIncomplete = false;
if (outputPtr > 0) {
outputCount += outputPtr;
out.write(buffer, 0, outputPtr);
}
return outputCount;
}
/*
/**********************************************************
/* Public API, traversal, basic
/**********************************************************
*/
Returns: Next token from the stream, if any found, or null
to indicate end-of-input
/**
* @return Next token from the stream, if any found, or null
* to indicate end-of-input
*/
@Override
public JsonToken nextToken() throws IOException
{
/* First: field names are special -- we will always tokenize
* (part of) value along with field name to simplify
* state handling. If so, can and need to use secondary token:
*/
if (_currToken == JsonToken.FIELD_NAME) {
return _nextAfterName();
}
// But if we didn't already have a name, and (partially?) decode number,
// need to ensure no numeric information is leaked
_numTypesValid = NR_UNKNOWN;
if (_tokenIncomplete) {
_skipString(); // only strings can be partial
}
int i = _skipWSOrEnd();
if (i < 0) { // end-of-input
// Close/release things like input source, symbol table and recyclable buffers
close();
return (_currToken = null);
}
// clear any data retained so far
_binaryValue = null;
// Closing scope?
if (i == INT_RBRACKET) {
_closeArrayScope();
return (_currToken = JsonToken.END_ARRAY);
}
if (i == INT_RCURLY) {
_closeObjectScope();
return (_currToken = JsonToken.END_OBJECT);
}
// Nope: do we then expect a comma?
if (_parsingContext.expectComma()) {
if (i != INT_COMMA) {
_reportUnexpectedChar(i, "was expecting comma to separate "+_parsingContext.typeDesc()+" entries");
}
i = _skipWS();
// Was that a trailing comma?
if ((_features & FEAT_MASK_TRAILING_COMMA) != 0) {
if ((i == INT_RBRACKET) || (i == INT_RCURLY)) {
return _closeScope(i);
}
}
}
/* And should we now have a name? Always true for Object contexts
* since the intermediate 'expect-value' state is never retained.
*/
if (!_parsingContext.inObject()) {
_updateLocation();
return _nextTokenNotInObject(i);
}
// So first parse the field name itself:
_updateNameLocation();
String n = _parseName(i);
_parsingContext.setCurrentName(n);
_currToken = JsonToken.FIELD_NAME;
i = _skipColon();
_updateLocation();
// Ok: we must have a value... what is it? Strings are very common, check first:
if (i == INT_QUOTE) {
_tokenIncomplete = true;
_nextToken = JsonToken.VALUE_STRING;
return _currToken;
}
JsonToken t;
switch (i) {
case '-':
t = _parseNegNumber();
break;
// Should we have separate handling for plus? Although it is not allowed per se,
// it may be erroneously used, and could be indicate by a more specific error message.
case '.': // [core#611]:
t = _parseFloatThatStartsWithPeriod();
break;
case '0':
case '1':
case '2':
case '3':
case '4':
case '5':
case '6':
case '7':
case '8':
case '9':
t = _parsePosNumber(i);
break;
case 'f':
_matchFalse();
t = JsonToken.VALUE_FALSE;
break;
case 'n':
_matchNull();
t = JsonToken.VALUE_NULL;
break;
case 't':
_matchTrue();
t = JsonToken.VALUE_TRUE;
break;
case '[':
t = JsonToken.START_ARRAY;
break;
case '{':
t = JsonToken.START_OBJECT;
break;
default:
t = _handleUnexpectedValue(i);
}
_nextToken = t;
return _currToken;
}
private final JsonToken _nextTokenNotInObject(int i) throws IOException
{
if (i == INT_QUOTE) {
_tokenIncomplete = true;
return (_currToken = JsonToken.VALUE_STRING);
}
switch (i) {
case '[':
_parsingContext = _parsingContext.createChildArrayContext(_tokenInputRow, _tokenInputCol);
return (_currToken = JsonToken.START_ARRAY);
case '{':
_parsingContext = _parsingContext.createChildObjectContext(_tokenInputRow, _tokenInputCol);
return (_currToken = JsonToken.START_OBJECT);
case 't':
_matchTrue();
return (_currToken = JsonToken.VALUE_TRUE);
case 'f':
_matchFalse();
return (_currToken = JsonToken.VALUE_FALSE);
case 'n':
_matchNull();
return (_currToken = JsonToken.VALUE_NULL);
case '-':
return (_currToken = _parseNegNumber());
// Should we have separate handling for plus? Although it is not allowed per se,
// it may be erroneously used, and could be indicate by a more specific error message.
case '.': // [core#611]:
return (_currToken = _parseFloatThatStartsWithPeriod());
case '0':
case '1':
case '2':
case '3':
case '4':
case '5':
case '6':
case '7':
case '8':
case '9':
return (_currToken = _parsePosNumber(i));
}
return (_currToken = _handleUnexpectedValue(i));
}
private final JsonToken _nextAfterName()
{
_nameCopied = false; // need to invalidate if it was copied
JsonToken t = _nextToken;
_nextToken = null;
// !!! 16-Nov-2015, tatu: TODO: fix [databind#37], copy next location to current here
// Also: may need to start new context?
if (t == JsonToken.START_ARRAY) {
_parsingContext = _parsingContext.createChildArrayContext(_tokenInputRow, _tokenInputCol);
} else if (t == JsonToken.START_OBJECT) {
_parsingContext = _parsingContext.createChildObjectContext(_tokenInputRow, _tokenInputCol);
}
return (_currToken = t);
}
@Override
public void finishToken() throws IOException {
if (_tokenIncomplete) {
_tokenIncomplete = false;
_finishString(); // only strings can be incomplete
}
}
/*
/**********************************************************
/* Public API, traversal, nextXxxValue/nextFieldName
/**********************************************************
*/
@Override
public boolean nextFieldName(SerializableString str) throws IOException
{
// // // Note: most of code below is copied from nextToken()
_numTypesValid = NR_UNKNOWN;
if (_currToken == JsonToken.FIELD_NAME) { // can't have name right after name
_nextAfterName();
return false;
}
if (_tokenIncomplete) {
_skipString();
}
int i = _skipWSOrEnd();
if (i < 0) { // end-of-input
close();
_currToken = null;
return false;
}
_binaryValue = null;
// Closing scope?
if (i == INT_RBRACKET) {
_closeArrayScope();
_currToken = JsonToken.END_ARRAY;
return false;
}
if (i == INT_RCURLY) {
_closeObjectScope();
_currToken = JsonToken.END_OBJECT;
return false;
}
// Nope: do we then expect a comma?
if (_parsingContext.expectComma()) {
if (i != INT_COMMA) {
_reportUnexpectedChar(i, "was expecting comma to separate "+_parsingContext.typeDesc()+" entries");
}
i = _skipWS();
// Was that a trailing comma?
if ((_features & FEAT_MASK_TRAILING_COMMA) != 0) {
if ((i == INT_RBRACKET) || (i == INT_RCURLY)) {
_closeScope(i);
return false;
}
}
}
if (!_parsingContext.inObject()) {
_updateLocation();
_nextTokenNotInObject(i);
return false;
}
// // // This part differs, name parsing
_updateNameLocation();
if (i == INT_QUOTE) {
// when doing literal match, must consider escaping:
byte[] nameBytes = str.asQuotedUTF8();
final int len = nameBytes.length;
// 22-May-2014, tatu: Actually, let's require 4 more bytes for faster skipping
// of colon that follows name
if ((_inputPtr + len + 4) < _inputEnd) { // maybe...
// first check length match by
final int end = _inputPtr+len;
if (_inputBuffer[end] == INT_QUOTE) {
int offset = 0;
int ptr = _inputPtr;
while (true) {
if (ptr == end) { // yes, match!
_parsingContext.setCurrentName(str.getValue());
i = _skipColonFast(ptr+1);
_isNextTokenNameYes(i);
return true;
}
if (nameBytes[offset] != _inputBuffer[ptr]) {
break;
}
++offset;
++ptr;
}
}
}
}
return _isNextTokenNameMaybe(i, str);
}
@Override
public String nextFieldName() throws IOException
{
// // // Note: this is almost a verbatim copy of nextToken()
_numTypesValid = NR_UNKNOWN;
if (_currToken == JsonToken.FIELD_NAME) {
_nextAfterName();
return null;
}
if (_tokenIncomplete) {
_skipString();
}
int i = _skipWSOrEnd();
if (i < 0) {
close();
_currToken = null;
return null;
}
_binaryValue = null;
if (i == INT_RBRACKET) {
_closeArrayScope();
_currToken = JsonToken.END_ARRAY;
return null;
}
if (i == INT_RCURLY) {
_closeObjectScope();
_currToken = JsonToken.END_OBJECT;
return null;
}
// Nope: do we then expect a comma?
if (_parsingContext.expectComma()) {
if (i != INT_COMMA) {
_reportUnexpectedChar(i, "was expecting comma to separate "+_parsingContext.typeDesc()+" entries");
}
i = _skipWS();
// Was that a trailing comma?
if ((_features & FEAT_MASK_TRAILING_COMMA) != 0) {
if ((i == INT_RBRACKET) || (i == INT_RCURLY)) {
_closeScope(i);
return null;
}
}
}
if (!_parsingContext.inObject()) {
_updateLocation();
_nextTokenNotInObject(i);
return null;
}
_updateNameLocation();
final String nameStr = _parseName(i);
_parsingContext.setCurrentName(nameStr);
_currToken = JsonToken.FIELD_NAME;
i = _skipColon();
_updateLocation();
if (i == INT_QUOTE) {
_tokenIncomplete = true;
_nextToken = JsonToken.VALUE_STRING;
return nameStr;
}
JsonToken t;
switch (i) {
case '-':
t = _parseNegNumber();
break;
case '.': // [core#611]:
t = _parseFloatThatStartsWithPeriod();
break;
case '0':
case '1':
case '2':
case '3':
case '4':
case '5':
case '6':
case '7':
case '8':
case '9':
t = _parsePosNumber(i);
break;
case 'f':
_matchFalse();
t = JsonToken.VALUE_FALSE;
break;
case 'n':
_matchNull();
t = JsonToken.VALUE_NULL;
break;
case 't':
_matchTrue();
t = JsonToken.VALUE_TRUE;
break;
case '[':
t = JsonToken.START_ARRAY;
break;
case '{':
t = JsonToken.START_OBJECT;
break;
default:
t = _handleUnexpectedValue(i);
}
_nextToken = t;
return nameStr;
}
// Variant called when we know there's at least 4 more bytes available
private final int _skipColonFast(int ptr) throws IOException
{
int i = _inputBuffer[ptr++];
if (i == INT_COLON) { // common case, no leading space
i = _inputBuffer[ptr++];
if (i > INT_SPACE) { // nor trailing
if (i != INT_SLASH && i != INT_HASH) {
_inputPtr = ptr;
return i;
}
} else if (i == INT_SPACE || i == INT_TAB) {
i = (int) _inputBuffer[ptr++];
if (i > INT_SPACE) {
if (i != INT_SLASH && i != INT_HASH) {
_inputPtr = ptr;
return i;
}
}
}
_inputPtr = ptr-1;
return _skipColon2(true); // true -> skipped colon
}
if (i == INT_SPACE || i == INT_TAB) {
i = _inputBuffer[ptr++];
}
if (i == INT_COLON) {
i = _inputBuffer[ptr++];
if (i > INT_SPACE) {
if (i != INT_SLASH && i != INT_HASH) {
_inputPtr = ptr;
return i;
}
} else if (i == INT_SPACE || i == INT_TAB) {
i = (int) _inputBuffer[ptr++];
if (i > INT_SPACE) {
if (i != INT_SLASH && i != INT_HASH) {
_inputPtr = ptr;
return i;
}
}
}
_inputPtr = ptr-1;
return _skipColon2(true);
}
_inputPtr = ptr-1;
return _skipColon2(false);
}
private final void _isNextTokenNameYes(int i) throws IOException
{
_currToken = JsonToken.FIELD_NAME;
_updateLocation();
switch (i) {
case '"':
_tokenIncomplete = true;
_nextToken = JsonToken.VALUE_STRING;
return;
case '[':
_nextToken = JsonToken.START_ARRAY;
return;
case '{':
_nextToken = JsonToken.START_OBJECT;
return;
case 't':
_matchTrue();
_nextToken = JsonToken.VALUE_TRUE;
return;
case 'f':
_matchFalse();
_nextToken = JsonToken.VALUE_FALSE;
return;
case 'n':
_matchNull();
_nextToken = JsonToken.VALUE_NULL;
return;
case '-':
_nextToken = _parseNegNumber();
return;
case '.': // [core#611]:
_nextToken = _parseFloatThatStartsWithPeriod();
return;
case '0':
case '1':
case '2':
case '3':
case '4':
case '5':
case '6':
case '7':
case '8':
case '9':
_nextToken = _parsePosNumber(i);
return;
}
_nextToken = _handleUnexpectedValue(i);
}
private final boolean _isNextTokenNameMaybe(int i, SerializableString str) throws IOException
{
// // // and this is back to standard nextToken()
String n = _parseName(i);
_parsingContext.setCurrentName(n);
final boolean match = n.equals(str.getValue());
_currToken = JsonToken.FIELD_NAME;
i = _skipColon();
_updateLocation();
// Ok: we must have a value... what is it? Strings are very common, check first:
if (i == INT_QUOTE) {
_tokenIncomplete = true;
_nextToken = JsonToken.VALUE_STRING;
return match;
}
JsonToken t;
switch (i) {
case '[':
t = JsonToken.START_ARRAY;
break;
case '{':
t = JsonToken.START_OBJECT;
break;
case 't':
_matchTrue();
t = JsonToken.VALUE_TRUE;
break;
case 'f':
_matchFalse();
t = JsonToken.VALUE_FALSE;
break;
case 'n':
_matchNull();
t = JsonToken.VALUE_NULL;
break;
case '-':
t = _parseNegNumber();
break;
case '.': // [core#611]:
t = _parseFloatThatStartsWithPeriod();
break;
case '0':
case '1':
case '2':
case '3':
case '4':
case '5':
case '6':
case '7':
case '8':
case '9':
t = _parsePosNumber(i);
break;
default:
t = _handleUnexpectedValue(i);
}
_nextToken = t;
return match;
}
@Override
public String nextTextValue() throws IOException
{
// two distinct cases; either got name and we know next type, or 'other'
if (_currToken == JsonToken.FIELD_NAME) { // mostly copied from '_nextAfterName'
_nameCopied = false;
JsonToken t = _nextToken;
_nextToken = null;
_currToken = t;
if (t == JsonToken.VALUE_STRING) {
if (_tokenIncomplete) {
_tokenIncomplete = false;
return _finishAndReturnString();
}
return _textBuffer.contentsAsString();
}
if (t == JsonToken.START_ARRAY) {
_parsingContext = _parsingContext.createChildArrayContext(_tokenInputRow, _tokenInputCol);
} else if (t == JsonToken.START_OBJECT) {
_parsingContext = _parsingContext.createChildObjectContext(_tokenInputRow, _tokenInputCol);
}
return null;
}
// !!! TODO: optimize this case as well
return (nextToken() == JsonToken.VALUE_STRING) ? getText() : null;
}
@Override
public int nextIntValue(int defaultValue) throws IOException
{
// two distinct cases; either got name and we know next type, or 'other'
if (_currToken == JsonToken.FIELD_NAME) { // mostly copied from '_nextAfterName'
_nameCopied = false;
JsonToken t = _nextToken;
_nextToken = null;
_currToken = t;
if (t == JsonToken.VALUE_NUMBER_INT) {
return getIntValue();
}
if (t == JsonToken.START_ARRAY) {
_parsingContext = _parsingContext.createChildArrayContext(_tokenInputRow, _tokenInputCol);
} else if (t == JsonToken.START_OBJECT) {
_parsingContext = _parsingContext.createChildObjectContext(_tokenInputRow, _tokenInputCol);
}
return defaultValue;
}
// !!! TODO: optimize this case as well
return (nextToken() == JsonToken.VALUE_NUMBER_INT) ? getIntValue() : defaultValue;
}
@Override
public long nextLongValue(long defaultValue) throws IOException
{
// two distinct cases; either got name and we know next type, or 'other'
if (_currToken == JsonToken.FIELD_NAME) { // mostly copied from '_nextAfterName'
_nameCopied = false;
JsonToken t = _nextToken;
_nextToken = null;
_currToken = t;
if (t == JsonToken.VALUE_NUMBER_INT) {
return getLongValue();
}
if (t == JsonToken.START_ARRAY) {
_parsingContext = _parsingContext.createChildArrayContext(_tokenInputRow, _tokenInputCol);
} else if (t == JsonToken.START_OBJECT) {
_parsingContext = _parsingContext.createChildObjectContext(_tokenInputRow, _tokenInputCol);
}
return defaultValue;
}
// !!! TODO: optimize this case as well
return (nextToken() == JsonToken.VALUE_NUMBER_INT) ? getLongValue() : defaultValue;
}
@Override
public Boolean nextBooleanValue() throws IOException
{
// two distinct cases; either got name and we know next type, or 'other'
if (_currToken == JsonToken.FIELD_NAME) { // mostly copied from '_nextAfterName'
_nameCopied = false;
JsonToken t = _nextToken;
_nextToken = null;
_currToken = t;
if (t == JsonToken.VALUE_TRUE) {
return Boolean.TRUE;
}
if (t == JsonToken.VALUE_FALSE) {
return Boolean.FALSE;
}
if (t == JsonToken.START_ARRAY) {
_parsingContext = _parsingContext.createChildArrayContext(_tokenInputRow, _tokenInputCol);
} else if (t == JsonToken.START_OBJECT) {
_parsingContext = _parsingContext.createChildObjectContext(_tokenInputRow, _tokenInputCol);
}
return null;
}
JsonToken t = nextToken();
if (t == JsonToken.VALUE_TRUE) {
return Boolean.TRUE;
}
if (t == JsonToken.VALUE_FALSE) {
return Boolean.FALSE;
}
return null;
}
/*
/**********************************************************
/* Internal methods, number parsing
/**********************************************************
*/
// @since 2.11, [core#611]
protected final JsonToken _parseFloatThatStartsWithPeriod() throws IOException
{
// [core#611]: allow optionally leading decimal point
if (!isEnabled(JsonReadFeature.ALLOW_LEADING_DECIMAL_POINT_FOR_NUMBERS.mappedFeature())) {
return _handleUnexpectedValue(INT_PERIOD);
}
return _parseFloat(_textBuffer.emptyAndGetCurrentSegment(),
0, INT_PERIOD, false, 0);
}
Initial parsing method for number values. It needs to be able
to parse enough input to be able to determine whether the
value is to be considered a simple integer value, or a more
generic decimal value: latter of which needs to be expressed
as a floating point number. The basic rule is that if the number
has no fractional or exponential part, it is an integer; otherwise
a floating point number.
Because much of input has to be processed in any case, no partial
parsing is done: all input text will be stored for further
processing. However, actual numeric value conversion will be
deferred, since it is usually the most complicated and costliest
part of processing.
/**
* Initial parsing method for number values. It needs to be able
* to parse enough input to be able to determine whether the
* value is to be considered a simple integer value, or a more
* generic decimal value: latter of which needs to be expressed
* as a floating point number. The basic rule is that if the number
* has no fractional or exponential part, it is an integer; otherwise
* a floating point number.
*<p>
* Because much of input has to be processed in any case, no partial
* parsing is done: all input text will be stored for further
* processing. However, actual numeric value conversion will be
* deferred, since it is usually the most complicated and costliest
* part of processing.
*/
protected JsonToken _parsePosNumber(int c) throws IOException
{
char[] outBuf = _textBuffer.emptyAndGetCurrentSegment();
// One special case: if first char is 0, must not be followed by a digit
if (c == INT_0) {
c = _verifyNoLeadingZeroes();
}
// Ok: we can first just add digit we saw first:
outBuf[0] = (char) c;
int intLen = 1;
int outPtr = 1;
// And then figure out how far we can read without further checks
// for either input or output
final int end = Math.min(_inputEnd, _inputPtr + outBuf.length - 1); // 1 == outPtr
// With this, we have a nice and tight loop:
while (true) {
if (_inputPtr >= end) { // split across boundary, offline
return _parseNumber2(outBuf, outPtr, false, intLen);
}
c = (int) _inputBuffer[_inputPtr++] & 0xFF;
if (c < INT_0 || c > INT_9) {
break;
}
++intLen;
outBuf[outPtr++] = (char) c;
}
if (c == INT_PERIOD || c == INT_e || c == INT_E) {
return _parseFloat(outBuf, outPtr, c, false, intLen);
}
--_inputPtr; // to push back trailing char (comma etc)
_textBuffer.setCurrentLength(outPtr);
// As per #105, need separating space between root values; check here
if (_parsingContext.inRoot()) {
_verifyRootSpace(c);
}
// And there we have it!
return resetInt(false, intLen);
}
protected JsonToken _parseNegNumber() throws IOException
{
char[] outBuf = _textBuffer.emptyAndGetCurrentSegment();
int outPtr = 0;
// Need to prepend sign?
outBuf[outPtr++] = '-';
// Must have something after sign too
if (_inputPtr >= _inputEnd) {
_loadMoreGuaranteed();
}
int c = (int) _inputBuffer[_inputPtr++] & 0xFF;
// Note: must be followed by a digit
if (c <= INT_0) {
// One special case: if first char is 0, must not be followed by a digit
if (c != INT_0) {
return _handleInvalidNumberStart(c, true);
}
c = _verifyNoLeadingZeroes();
} else if (c > INT_9) {
return _handleInvalidNumberStart(c, true);
}
// Ok: we can first just add digit we saw first:
outBuf[outPtr++] = (char) c;
int intLen = 1;
// And then figure out how far we can read without further checks
// for either input or output
final int end = Math.min(_inputEnd, _inputPtr + outBuf.length - outPtr);
// With this, we have a nice and tight loop:
while (true) {
if (_inputPtr >= end) {
// Long enough to be split across boundary, so:
return _parseNumber2(outBuf, outPtr, true, intLen);
}
c = (int) _inputBuffer[_inputPtr++] & 0xFF;
if (c < INT_0 || c > INT_9) {
break;
}
++intLen;
outBuf[outPtr++] = (char) c;
}
if (c == INT_PERIOD || c == INT_e || c == INT_E) {
return _parseFloat(outBuf, outPtr, c, true, intLen);
}
--_inputPtr; // to push back trailing char (comma etc)
_textBuffer.setCurrentLength(outPtr);
// As per #105, need separating space between root values; check here
if (_parsingContext.inRoot()) {
_verifyRootSpace(c);
}
// And there we have it!
return resetInt(true, intLen);
}
Method called to handle parsing when input is split across buffer boundary
(or output is longer than segment used to store it)
/**
* Method called to handle parsing when input is split across buffer boundary
* (or output is longer than segment used to store it)
*/
private final JsonToken _parseNumber2(char[] outBuf, int outPtr, boolean negative,
int intPartLength) throws IOException
{
// Ok, parse the rest
while (true) {
if (_inputPtr >= _inputEnd && !_loadMore()) {
_textBuffer.setCurrentLength(outPtr);
return resetInt(negative, intPartLength);
}
int c = (int) _inputBuffer[_inputPtr++] & 0xFF;
if (c > INT_9 || c < INT_0) {
if (c == INT_PERIOD || c == INT_e || c == INT_E) {
return _parseFloat(outBuf, outPtr, c, negative, intPartLength);
}
break;
}
if (outPtr >= outBuf.length) {
outBuf = _textBuffer.finishCurrentSegment();
outPtr = 0;
}
outBuf[outPtr++] = (char) c;
++intPartLength;
}
--_inputPtr; // to push back trailing char (comma etc)
_textBuffer.setCurrentLength(outPtr);
// As per #105, need separating space between root values; check here
if (_parsingContext.inRoot()) {
_verifyRootSpace(_inputBuffer[_inputPtr] & 0xFF);
}
// And there we have it!
return resetInt(negative, intPartLength);
}
Method called when we have seen one zero, and want to ensure
it is not followed by another
/**
* Method called when we have seen one zero, and want to ensure
* it is not followed by another
*/
private final int _verifyNoLeadingZeroes() throws IOException
{
// Ok to have plain "0"
if (_inputPtr >= _inputEnd && !_loadMore()) {
return INT_0;
}
int ch = _inputBuffer[_inputPtr] & 0xFF;
// if not followed by a number (probably '.'); return zero as is, to be included
if (ch < INT_0 || ch > INT_9) {
return INT_0;
}
// [JACKSON-358]: we may want to allow them, after all...
if ((_features & FEAT_MASK_LEADING_ZEROS) == 0) {
reportInvalidNumber("Leading zeroes not allowed");
}
// if so, just need to skip either all zeroes (if followed by number); or all but one (if non-number)
++_inputPtr; // Leading zero to be skipped
if (ch == INT_0) {
while (_inputPtr < _inputEnd || _loadMore()) {
ch = _inputBuffer[_inputPtr] & 0xFF;
if (ch < INT_0 || ch > INT_9) { // followed by non-number; retain one zero
return INT_0;
}
++_inputPtr; // skip previous zeroes
if (ch != INT_0) { // followed by other number; return
break;
}
}
}
return ch;
}
private final JsonToken _parseFloat(char[] outBuf, int outPtr, int c,
boolean negative, int integerPartLength) throws IOException
{
int fractLen = 0;
boolean eof = false;
// And then see if we get other parts
if (c == INT_PERIOD) { // yes, fraction
if (outPtr >= outBuf.length) {
outBuf = _textBuffer.finishCurrentSegment();
outPtr = 0;
}
outBuf[outPtr++] = (char) c;
fract_loop:
while (true) {
if (_inputPtr >= _inputEnd && !_loadMore()) {
eof = true;
break fract_loop;
}
c = (int) _inputBuffer[_inputPtr++] & 0xFF;
if (c < INT_0 || c > INT_9) {
break fract_loop;
}
++fractLen;
if (outPtr >= outBuf.length) {
outBuf = _textBuffer.finishCurrentSegment();
outPtr = 0;
}
outBuf[outPtr++] = (char) c;
}
// must be followed by sequence of ints, one minimum
if (fractLen == 0) {
reportUnexpectedNumberChar(c, "Decimal point not followed by a digit");
}
}
int expLen = 0;
if (c == INT_e || c == INT_E) { // exponent?
if (outPtr >= outBuf.length) {
outBuf = _textBuffer.finishCurrentSegment();
outPtr = 0;
}
outBuf[outPtr++] = (char) c;
// Not optional, can require that we get one more char
if (_inputPtr >= _inputEnd) {
_loadMoreGuaranteed();
}
c = (int) _inputBuffer[_inputPtr++] & 0xFF;
// Sign indicator?
if (c == '-' || c == '+') {
if (outPtr >= outBuf.length) {
outBuf = _textBuffer.finishCurrentSegment();
outPtr = 0;
}
outBuf[outPtr++] = (char) c;
// Likewise, non optional:
if (_inputPtr >= _inputEnd) {
_loadMoreGuaranteed();
}
c = (int) _inputBuffer[_inputPtr++] & 0xFF;
}
exp_loop:
while (c >= INT_0 && c <= INT_9) {
++expLen;
if (outPtr >= outBuf.length) {
outBuf = _textBuffer.finishCurrentSegment();
outPtr = 0;
}
outBuf[outPtr++] = (char) c;
if (_inputPtr >= _inputEnd && !_loadMore()) {
eof = true;
break exp_loop;
}
c = (int) _inputBuffer[_inputPtr++] & 0xFF;
}
// must be followed by sequence of ints, one minimum
if (expLen == 0) {
reportUnexpectedNumberChar(c, "Exponent indicator not followed by a digit");
}
}
// Ok; unless we hit end-of-input, need to push last char read back
if (!eof) {
--_inputPtr;
// As per [core#105], need separating space between root values; check here
if (_parsingContext.inRoot()) {
_verifyRootSpace(c);
}
}
_textBuffer.setCurrentLength(outPtr);
// And there we have it!
return resetFloat(negative, integerPartLength, fractLen, expLen);
}
Method called to ensure that a root-value is followed by a space
token.
NOTE: caller MUST ensure there is at least one character available;
and that input pointer is AT given char (not past)
/**
* Method called to ensure that a root-value is followed by a space
* token.
*<p>
* NOTE: caller MUST ensure there is at least one character available;
* and that input pointer is AT given char (not past)
*/
private final void _verifyRootSpace(int ch) throws IOException
{
// caller had pushed it back, before calling; reset
++_inputPtr;
// TODO? Handle UTF-8 char decoding for error reporting
switch (ch) {
case ' ':
case '\t':
return;
case '\r':
_skipCR();
return;
case '\n':
++_currInputRow;
_currInputRowStart = _inputPtr;
return;
}
_reportMissingRootWS(ch);
}
/*
/**********************************************************
/* Internal methods, secondary parsing
/**********************************************************
*/
protected final String _parseName(int i) throws IOException
{
if (i != INT_QUOTE) {
return _handleOddName(i);
}
// First: can we optimize out bounds checks?
if ((_inputPtr + 13) > _inputEnd) { // Need up to 12 chars, plus one trailing (quote)
return slowParseName();
}
// If so, can also unroll loops nicely
/* 25-Nov-2008, tatu: This may seem weird, but here we do
* NOT want to worry about UTF-8 decoding. Rather, we'll
* assume that part is ok (if not it will get caught
* later on), and just handle quotes and backslashes here.
*/
final byte[] input = _inputBuffer;
final int[] codes = _icLatin1;
int q = input[_inputPtr++] & 0xFF;
if (codes[q] == 0) {
i = input[_inputPtr++] & 0xFF;
if (codes[i] == 0) {
q = (q << 8) | i;
i = input[_inputPtr++] & 0xFF;
if (codes[i] == 0) {
q = (q << 8) | i;
i = input[_inputPtr++] & 0xFF;
if (codes[i] == 0) {
q = (q << 8) | i;
i = input[_inputPtr++] & 0xFF;
if (codes[i] == 0) {
_quad1 = q;
return parseMediumName(i);
}
if (i == INT_QUOTE) { // 4 byte/char case or broken
return findName(q, 4);
}
return parseName(q, i, 4);
}
if (i == INT_QUOTE) { // 3 byte/char case or broken
return findName(q, 3);
}
return parseName(q, i, 3);
}
if (i == INT_QUOTE) { // 2 byte/char case or broken
return findName(q, 2);
}
return parseName(q, i, 2);
}
if (i == INT_QUOTE) { // one byte/char case or broken
return findName(q, 1);
}
return parseName(q, i, 1);
}
if (q == INT_QUOTE) { // special case, ""
return "";
}
return parseName(0, q, 0); // quoting or invalid char
}
protected final String parseMediumName(int q2) throws IOException
{
final byte[] input = _inputBuffer;
final int[] codes = _icLatin1;
// Ok, got 5 name bytes so far
int i = input[_inputPtr++] & 0xFF;
if (codes[i] != 0) {
if (i == INT_QUOTE) { // 5 bytes
return findName(_quad1, q2, 1);
}
return parseName(_quad1, q2, i, 1); // quoting or invalid char
}
q2 = (q2 << 8) | i;
i = input[_inputPtr++] & 0xFF;
if (codes[i] != 0) {
if (i == INT_QUOTE) { // 6 bytes
return findName(_quad1, q2, 2);
}
return parseName(_quad1, q2, i, 2);
}
q2 = (q2 << 8) | i;
i = input[_inputPtr++] & 0xFF;
if (codes[i] != 0) {
if (i == INT_QUOTE) { // 7 bytes
return findName(_quad1, q2, 3);
}
return parseName(_quad1, q2, i, 3);
}
q2 = (q2 << 8) | i;
i = input[_inputPtr++] & 0xFF;
if (codes[i] != 0) {
if (i == INT_QUOTE) { // 8 bytes
return findName(_quad1, q2, 4);
}
return parseName(_quad1, q2, i, 4);
}
return parseMediumName2(i, q2);
}
Since: 2.6
/**
* @since 2.6
*/
protected final String parseMediumName2(int q3, final int q2) throws IOException
{
final byte[] input = _inputBuffer;
final int[] codes = _icLatin1;
// Got 9 name bytes so far
int i = input[_inputPtr++] & 0xFF;
if (codes[i] != 0) {
if (i == INT_QUOTE) { // 9 bytes
return findName(_quad1, q2, q3, 1);
}
return parseName(_quad1, q2, q3, i, 1);
}
q3 = (q3 << 8) | i;
i = input[_inputPtr++] & 0xFF;
if (codes[i] != 0) {
if (i == INT_QUOTE) { // 10 bytes
return findName(_quad1, q2, q3, 2);
}
return parseName(_quad1, q2, q3, i, 2);
}
q3 = (q3 << 8) | i;
i = input[_inputPtr++] & 0xFF;
if (codes[i] != 0) {
if (i == INT_QUOTE) { // 11 bytes
return findName(_quad1, q2, q3, 3);
}
return parseName(_quad1, q2, q3, i, 3);
}
q3 = (q3 << 8) | i;
i = input[_inputPtr++] & 0xFF;
if (codes[i] != 0) {
if (i == INT_QUOTE) { // 12 bytes
return findName(_quad1, q2, q3, 4);
}
return parseName(_quad1, q2, q3, i, 4);
}
return parseLongName(i, q2, q3);
}
protected final String parseLongName(int q, final int q2, int q3) throws IOException
{
_quadBuffer[0] = _quad1;
_quadBuffer[1] = q2;
_quadBuffer[2] = q3;
// As explained above, will ignore UTF-8 encoding at this point
final byte[] input = _inputBuffer;
final int[] codes = _icLatin1;
int qlen = 3;
while ((_inputPtr + 4) <= _inputEnd) {
int i = input[_inputPtr++] & 0xFF;
if (codes[i] != 0) {
if (i == INT_QUOTE) {
return findName(_quadBuffer, qlen, q, 1);
}
return parseEscapedName(_quadBuffer, qlen, q, i, 1);
}
q = (q << 8) | i;
i = input[_inputPtr++] & 0xFF;
if (codes[i] != 0) {
if (i == INT_QUOTE) {
return findName(_quadBuffer, qlen, q, 2);
}
return parseEscapedName(_quadBuffer, qlen, q, i, 2);
}
q = (q << 8) | i;
i = input[_inputPtr++] & 0xFF;
if (codes[i] != 0) {
if (i == INT_QUOTE) {
return findName(_quadBuffer, qlen, q, 3);
}
return parseEscapedName(_quadBuffer, qlen, q, i, 3);
}
q = (q << 8) | i;
i = input[_inputPtr++] & 0xFF;
if (codes[i] != 0) {
if (i == INT_QUOTE) {
return findName(_quadBuffer, qlen, q, 4);
}
return parseEscapedName(_quadBuffer, qlen, q, i, 4);
}
// Nope, no end in sight. Need to grow quad array etc
if (qlen >= _quadBuffer.length) {
_quadBuffer = growArrayBy(_quadBuffer, qlen);
}
_quadBuffer[qlen++] = q;
q = i;
}
/* Let's offline if we hit buffer boundary (otherwise would
* need to [try to] align input, which is bit complicated
* and may not always be possible)
*/
return parseEscapedName(_quadBuffer, qlen, 0, q, 0);
}
Method called when not even first 8 bytes are guaranteed
to come consecutively. Happens rarely, so this is offlined;
plus we'll also do full checks for escaping etc.
/**
* Method called when not even first 8 bytes are guaranteed
* to come consecutively. Happens rarely, so this is offlined;
* plus we'll also do full checks for escaping etc.
*/
protected String slowParseName() throws IOException
{
if (_inputPtr >= _inputEnd) {
if (!_loadMore()) {
_reportInvalidEOF(": was expecting closing '\"' for name", JsonToken.FIELD_NAME);
}
}
int i = _inputBuffer[_inputPtr++] & 0xFF;
if (i == INT_QUOTE) { // special case, ""
return "";
}
return parseEscapedName(_quadBuffer, 0, 0, i, 0);
}
private final String parseName(int q1, int ch, int lastQuadBytes) throws IOException {
return parseEscapedName(_quadBuffer, 0, q1, ch, lastQuadBytes);
}
private final String parseName(int q1, int q2, int ch, int lastQuadBytes) throws IOException {
_quadBuffer[0] = q1;
return parseEscapedName(_quadBuffer, 1, q2, ch, lastQuadBytes);
}
private final String parseName(int q1, int q2, int q3, int ch, int lastQuadBytes) throws IOException {
_quadBuffer[0] = q1;
_quadBuffer[1] = q2;
return parseEscapedName(_quadBuffer, 2, q3, ch, lastQuadBytes);
}
Slower parsing method which is generally branched to when an escape
sequence is detected (or alternatively for long names, one crossing
input buffer boundary). Needs to be able to handle more exceptional
cases, gets slower, and hence is offlined to a separate method.
/**
* Slower parsing method which is generally branched to when an escape
* sequence is detected (or alternatively for long names, one crossing
* input buffer boundary). Needs to be able to handle more exceptional
* cases, gets slower, and hence is offlined to a separate method.
*/
protected final String parseEscapedName(int[] quads, int qlen, int currQuad, int ch,
int currQuadBytes) throws IOException
{
// This may seem weird, but here we do not want to worry about
// UTF-8 decoding yet. Rather, we'll assume that part is ok (if not it will get
// caught later on), and just handle quotes and backslashes here.
final int[] codes = _icLatin1;
while (true) {
if (codes[ch] != 0) {
if (ch == INT_QUOTE) { // we are done
break;
}
// Unquoted white space?
if (ch != INT_BACKSLASH) {
// As per [JACKSON-208], call can now return:
_throwUnquotedSpace(ch, "name");
} else {
// Nope, escape sequence
ch = _decodeEscaped();
}
// Oh crap. May need to UTF-8 (re-)encode it, if it's beyond
// 7-bit ASCII. Gets pretty messy. If this happens often, may
// want to use different name canonicalization to avoid these hits.
if (ch > 127) {
// Ok, we'll need room for first byte right away
if (currQuadBytes >= 4) {
if (qlen >= quads.length) {
_quadBuffer = quads = growArrayBy(quads, quads.length);
}
quads[qlen++] = currQuad;
currQuad = 0;
currQuadBytes = 0;
}
if (ch < 0x800) { // 2-byte
currQuad = (currQuad << 8) | (0xc0 | (ch >> 6));
++currQuadBytes;
// Second byte gets output below:
} else { // 3 bytes; no need to worry about surrogates here
currQuad = (currQuad << 8) | (0xe0 | (ch >> 12));
++currQuadBytes;
// need room for middle byte?
if (currQuadBytes >= 4) {
if (qlen >= quads.length) {
_quadBuffer = quads = growArrayBy(quads, quads.length);
}
quads[qlen++] = currQuad;
currQuad = 0;
currQuadBytes = 0;
}
currQuad = (currQuad << 8) | (0x80 | ((ch >> 6) & 0x3f));
++currQuadBytes;
}
// And same last byte in both cases, gets output below:
ch = 0x80 | (ch & 0x3f);
}
}
// Ok, we have one more byte to add at any rate:
if (currQuadBytes < 4) {
++currQuadBytes;
currQuad = (currQuad << 8) | ch;
} else {
if (qlen >= quads.length) {
_quadBuffer = quads = growArrayBy(quads, quads.length);
}
quads[qlen++] = currQuad;
currQuad = ch;
currQuadBytes = 1;
}
if (_inputPtr >= _inputEnd) {
if (!_loadMore()) {
_reportInvalidEOF(" in field name", JsonToken.FIELD_NAME);
}
}
ch = _inputBuffer[_inputPtr++] & 0xFF;
}
if (currQuadBytes > 0) {
if (qlen >= quads.length) {
_quadBuffer = quads = growArrayBy(quads, quads.length);
}
quads[qlen++] = _padLastQuad(currQuad, currQuadBytes);
}
String name = _symbols.findName(quads, qlen);
if (name == null) {
name = addName(quads, qlen, currQuadBytes);
}
return name;
}
Method called when we see non-white space character other
than double quote, when expecting a field name.
In standard mode will just throw an exception; but
in non-standard modes may be able to parse name.
/**
* Method called when we see non-white space character other
* than double quote, when expecting a field name.
* In standard mode will just throw an exception; but
* in non-standard modes may be able to parse name.
*/
protected String _handleOddName(int ch) throws IOException
{
// First: may allow single quotes
if (ch == INT_APOS && (_features & FEAT_MASK_ALLOW_SINGLE_QUOTES) != 0) {
return _parseAposName();
}
// Allow unquoted names if feature enabled:
if ((_features & FEAT_MASK_ALLOW_UNQUOTED_NAMES) == 0) {
char c = (char) _decodeCharForError(ch);
_reportUnexpectedChar(c, "was expecting double-quote to start field name");
}
/* Also: note that although we use a different table here,
* it does NOT handle UTF-8 decoding. It'll just pass those
* high-bit codes as acceptable for later decoding.
*/
final int[] codes = CharTypes.getInputCodeUtf8JsNames();
// Also: must start with a valid character...
if (codes[ch] != 0) {
_reportUnexpectedChar(ch, "was expecting either valid name character (for unquoted name) or double-quote (for quoted) to start field name");
}
// Ok, now; instead of ultra-optimizing parsing here (as with regular
// JSON names), let's just use the generic "slow" variant.
// Can measure its impact later on if need be.
int[] quads = _quadBuffer;
int qlen = 0;
int currQuad = 0;
int currQuadBytes = 0;
while (true) {
// Ok, we have one more byte to add at any rate:
if (currQuadBytes < 4) {
++currQuadBytes;
currQuad = (currQuad << 8) | ch;
} else {
if (qlen >= quads.length) {
_quadBuffer = quads = growArrayBy(quads, quads.length);
}
quads[qlen++] = currQuad;
currQuad = ch;
currQuadBytes = 1;
}
if (_inputPtr >= _inputEnd) {
if (!_loadMore()) {
_reportInvalidEOF(" in field name", JsonToken.FIELD_NAME);
}
}
ch = _inputBuffer[_inputPtr] & 0xFF;
if (codes[ch] != 0) {
break;
}
++_inputPtr;
}
if (currQuadBytes > 0) {
if (qlen >= quads.length) {
_quadBuffer = quads = growArrayBy(quads, quads.length);
}
quads[qlen++] = currQuad;
}
String name = _symbols.findName(quads, qlen);
if (name == null) {
name = addName(quads, qlen, currQuadBytes);
}
return name;
}
/* Parsing to support [JACKSON-173]. Plenty of duplicated code;
* main reason being to try to avoid slowing down fast path
* for valid JSON -- more alternatives, more code, generally
* bit slower execution.
*/
protected String _parseAposName() throws IOException
{
if (_inputPtr >= _inputEnd) {
if (!_loadMore()) {
_reportInvalidEOF(": was expecting closing '\'' for field name", JsonToken.FIELD_NAME);
}
}
int ch = _inputBuffer[_inputPtr++] & 0xFF;
if (ch == INT_APOS) { // special case, ''
return "";
}
int[] quads = _quadBuffer;
int qlen = 0;
int currQuad = 0;
int currQuadBytes = 0;
// Copied from parseEscapedFieldName, with minor mods:
final int[] codes = _icLatin1;
while (true) {
if (ch == INT_APOS) {
break;
}
// additional check to skip handling of double-quotes
if ((codes[ch] != 0) && (ch != '"')) {
if (ch != '\\') {
// Unquoted white space?
// As per [JACKSON-208], call can now return:
_throwUnquotedSpace(ch, "name");
} else {
// Nope, escape sequence
ch = _decodeEscaped();
}
// as per main code, inefficient but will have to do
if (ch > 127) {
// Ok, we'll need room for first byte right away
if (currQuadBytes >= 4) {
if (qlen >= quads.length) {
_quadBuffer = quads = growArrayBy(quads, quads.length);
}
quads[qlen++] = currQuad;
currQuad = 0;
currQuadBytes = 0;
}
if (ch < 0x800) { // 2-byte
currQuad = (currQuad << 8) | (0xc0 | (ch >> 6));
++currQuadBytes;
// Second byte gets output below:
} else { // 3 bytes; no need to worry about surrogates here
currQuad = (currQuad << 8) | (0xe0 | (ch >> 12));
++currQuadBytes;
// need room for middle byte?
if (currQuadBytes >= 4) {
if (qlen >= quads.length) {
_quadBuffer = quads = growArrayBy(quads, quads.length);
}
quads[qlen++] = currQuad;
currQuad = 0;
currQuadBytes = 0;
}
currQuad = (currQuad << 8) | (0x80 | ((ch >> 6) & 0x3f));
++currQuadBytes;
}
// And same last byte in both cases, gets output below:
ch = 0x80 | (ch & 0x3f);
}
}
// Ok, we have one more byte to add at any rate:
if (currQuadBytes < 4) {
++currQuadBytes;
currQuad = (currQuad << 8) | ch;
} else {
if (qlen >= quads.length) {
_quadBuffer = quads = growArrayBy(quads, quads.length);
}
quads[qlen++] = currQuad;
currQuad = ch;
currQuadBytes = 1;
}
if (_inputPtr >= _inputEnd) {
if (!_loadMore()) {
_reportInvalidEOF(" in field name", JsonToken.FIELD_NAME);
}
}
ch = _inputBuffer[_inputPtr++] & 0xFF;
}
if (currQuadBytes > 0) {
if (qlen >= quads.length) {
_quadBuffer = quads = growArrayBy(quads, quads.length);
}
quads[qlen++] = _padLastQuad(currQuad, currQuadBytes);
}
String name = _symbols.findName(quads, qlen);
if (name == null) {
name = addName(quads, qlen, currQuadBytes);
}
return name;
}
/*
/**********************************************************
/* Internal methods, symbol (name) handling
/**********************************************************
*/
private final String findName(int q1, int lastQuadBytes) throws JsonParseException
{
q1 = _padLastQuad(q1, lastQuadBytes);
// Usually we'll find it from the canonical symbol table already
String name = _symbols.findName(q1);
if (name != null) {
return name;
}
// If not, more work. We'll need add stuff to buffer
_quadBuffer[0] = q1;
return addName(_quadBuffer, 1, lastQuadBytes);
}
private final String findName(int q1, int q2, int lastQuadBytes) throws JsonParseException
{
q2 = _padLastQuad(q2, lastQuadBytes);
// Usually we'll find it from the canonical symbol table already
String name = _symbols.findName(q1, q2);
if (name != null) {
return name;
}
// If not, more work. We'll need add stuff to buffer
_quadBuffer[0] = q1;
_quadBuffer[1] = q2;
return addName(_quadBuffer, 2, lastQuadBytes);
}
private final String findName(int q1, int q2, int q3, int lastQuadBytes) throws JsonParseException
{
q3 = _padLastQuad(q3, lastQuadBytes);
String name = _symbols.findName(q1, q2, q3);
if (name != null) {
return name;
}
int[] quads = _quadBuffer;
quads[0] = q1;
quads[1] = q2;
quads[2] = _padLastQuad(q3, lastQuadBytes);
return addName(quads, 3, lastQuadBytes);
}
private final String findName(int[] quads, int qlen, int lastQuad, int lastQuadBytes) throws JsonParseException
{
if (qlen >= quads.length) {
_quadBuffer = quads = growArrayBy(quads, quads.length);
}
quads[qlen++] = _padLastQuad(lastQuad, lastQuadBytes);
String name = _symbols.findName(quads, qlen);
if (name == null) {
return addName(quads, qlen, lastQuadBytes);
}
return name;
}
This is the main workhorse method used when we take a symbol
table miss. It needs to demultiplex individual bytes, decode
multi-byte chars (if any), and then construct Name instance
and add it to the symbol table.
/**
* This is the main workhorse method used when we take a symbol
* table miss. It needs to demultiplex individual bytes, decode
* multi-byte chars (if any), and then construct Name instance
* and add it to the symbol table.
*/
private final String addName(int[] quads, int qlen, int lastQuadBytes) throws JsonParseException
{
/* Ok: must decode UTF-8 chars. No other validation is
* needed, since unescaping has been done earlier as necessary
* (as well as error reporting for unescaped control chars)
*/
// 4 bytes per quad, except last one maybe less
int byteLen = (qlen << 2) - 4 + lastQuadBytes;
/* And last one is not correctly aligned (leading zero bytes instead
* need to shift a bit, instead of trailing). Only need to shift it
* for UTF-8 decoding; need revert for storage (since key will not
* be aligned, to optimize lookup speed)
*/
int lastQuad;
if (lastQuadBytes < 4) {
lastQuad = quads[qlen-1];
// 8/16/24 bit left shift
quads[qlen-1] = (lastQuad << ((4 - lastQuadBytes) << 3));
} else {
lastQuad = 0;
}
// Need some working space, TextBuffer works well:
char[] cbuf = _textBuffer.emptyAndGetCurrentSegment();
int cix = 0;
for (int ix = 0; ix < byteLen; ) {
int ch = quads[ix >> 2]; // current quad, need to shift+mask
int byteIx = (ix & 3);
ch = (ch >> ((3 - byteIx) << 3)) & 0xFF;
++ix;
if (ch > 127) { // multi-byte
int needed;
if ((ch & 0xE0) == 0xC0) { // 2 bytes (0x0080 - 0x07FF)
ch &= 0x1F;
needed = 1;
} else if ((ch & 0xF0) == 0xE0) { // 3 bytes (0x0800 - 0xFFFF)
ch &= 0x0F;
needed = 2;
} else if ((ch & 0xF8) == 0xF0) { // 4 bytes; double-char with surrogates and all...
ch &= 0x07;
needed = 3;
} else { // 5- and 6-byte chars not valid json chars
_reportInvalidInitial(ch);
needed = ch = 1; // never really gets this far
}
if ((ix + needed) > byteLen) {
_reportInvalidEOF(" in field name", JsonToken.FIELD_NAME);
}
// Ok, always need at least one more:
int ch2 = quads[ix >> 2]; // current quad, need to shift+mask
byteIx = (ix & 3);
ch2 = (ch2 >> ((3 - byteIx) << 3));
++ix;
if ((ch2 & 0xC0) != 0x080) {
_reportInvalidOther(ch2);
}
ch = (ch << 6) | (ch2 & 0x3F);
if (needed > 1) {
ch2 = quads[ix >> 2];
byteIx = (ix & 3);
ch2 = (ch2 >> ((3 - byteIx) << 3));
++ix;
if ((ch2 & 0xC0) != 0x080) {
_reportInvalidOther(ch2);
}
ch = (ch << 6) | (ch2 & 0x3F);
if (needed > 2) { // 4 bytes? (need surrogates on output)
ch2 = quads[ix >> 2];
byteIx = (ix & 3);
ch2 = (ch2 >> ((3 - byteIx) << 3));
++ix;
if ((ch2 & 0xC0) != 0x080) {
_reportInvalidOther(ch2 & 0xFF);
}
ch = (ch << 6) | (ch2 & 0x3F);
}
}
if (needed > 2) { // surrogate pair? once again, let's output one here, one later on
ch -= 0x10000; // to normalize it starting with 0x0
if (cix >= cbuf.length) {
cbuf = _textBuffer.expandCurrentSegment();
}
cbuf[cix++] = (char) (0xD800 + (ch >> 10));
ch = 0xDC00 | (ch & 0x03FF);
}
}
if (cix >= cbuf.length) {
cbuf = _textBuffer.expandCurrentSegment();
}
cbuf[cix++] = (char) ch;
}
// Ok. Now we have the character array, and can construct the String
String baseName = new String(cbuf, 0, cix);
// And finally, un-align if necessary
if (lastQuadBytes < 4) {
quads[qlen-1] = lastQuad;
}
return _symbols.addName(baseName, quads, qlen);
}
Helper method needed to fix [jackson-core#148], masking of 0x00 character
/**
* Helper method needed to fix [jackson-core#148], masking of 0x00 character
*/
private final static int _padLastQuad(int q, int bytes) {
return (bytes == 4) ? q : (q | (-1 << (bytes << 3)));
}
/*
/**********************************************************
/* Internal methods, String value parsing
/**********************************************************
*/
protected void _loadMoreGuaranteed() throws IOException {
if (!_loadMore()) { _reportInvalidEOF(); }
}
@Override
protected void _finishString() throws IOException
{
// First, single tight loop for ASCII content, not split across input buffer boundary:
int ptr = _inputPtr;
if (ptr >= _inputEnd) {
_loadMoreGuaranteed();
ptr = _inputPtr;
}
int outPtr = 0;
char[] outBuf = _textBuffer.emptyAndGetCurrentSegment();
final int[] codes = _icUTF8;
final int max = Math.min(_inputEnd, (ptr + outBuf.length));
final byte[] inputBuffer = _inputBuffer;
while (ptr < max) {
int c = (int) inputBuffer[ptr] & 0xFF;
if (codes[c] != 0) {
if (c == INT_QUOTE) {
_inputPtr = ptr+1;
_textBuffer.setCurrentLength(outPtr);
return;
}
break;
}
++ptr;
outBuf[outPtr++] = (char) c;
}
_inputPtr = ptr;
_finishString2(outBuf, outPtr);
}
Since: 2.6
/**
* @since 2.6
*/
protected String _finishAndReturnString() throws IOException
{
// First, single tight loop for ASCII content, not split across input buffer boundary:
int ptr = _inputPtr;
if (ptr >= _inputEnd) {
_loadMoreGuaranteed();
ptr = _inputPtr;
}
int outPtr = 0;
char[] outBuf = _textBuffer.emptyAndGetCurrentSegment();
final int[] codes = _icUTF8;
final int max = Math.min(_inputEnd, (ptr + outBuf.length));
final byte[] inputBuffer = _inputBuffer;
while (ptr < max) {
int c = (int) inputBuffer[ptr] & 0xFF;
if (codes[c] != 0) {
if (c == INT_QUOTE) {
_inputPtr = ptr+1;
return _textBuffer.setCurrentAndReturn(outPtr);
}
break;
}
++ptr;
outBuf[outPtr++] = (char) c;
}
_inputPtr = ptr;
_finishString2(outBuf, outPtr);
return _textBuffer.contentsAsString();
}
private final void _finishString2(char[] outBuf, int outPtr)
throws IOException
{
int c;
// Here we do want to do full decoding, hence:
final int[] codes = _icUTF8;
final byte[] inputBuffer = _inputBuffer;
main_loop:
while (true) {
// Then the tight ASCII non-funny-char loop:
ascii_loop:
while (true) {
int ptr = _inputPtr;
if (ptr >= _inputEnd) {
_loadMoreGuaranteed();
ptr = _inputPtr;
}
if (outPtr >= outBuf.length) {
outBuf = _textBuffer.finishCurrentSegment();
outPtr = 0;
}
final int max = Math.min(_inputEnd, (ptr + (outBuf.length - outPtr)));
while (ptr < max) {
c = (int) inputBuffer[ptr++] & 0xFF;
if (codes[c] != 0) {
_inputPtr = ptr;
break ascii_loop;
}
outBuf[outPtr++] = (char) c;
}
_inputPtr = ptr;
}
// Ok: end marker, escape or multi-byte?
if (c == INT_QUOTE) {
break main_loop;
}
switch (codes[c]) {
case 1: // backslash
c = _decodeEscaped();
break;
case 2: // 2-byte UTF
c = _decodeUtf8_2(c);
break;
case 3: // 3-byte UTF
if ((_inputEnd - _inputPtr) >= 2) {
c = _decodeUtf8_3fast(c);
} else {
c = _decodeUtf8_3(c);
}
break;
case 4: // 4-byte UTF
c = _decodeUtf8_4(c);
// Let's add first part right away:
outBuf[outPtr++] = (char) (0xD800 | (c >> 10));
if (outPtr >= outBuf.length) {
outBuf = _textBuffer.finishCurrentSegment();
outPtr = 0;
}
c = 0xDC00 | (c & 0x3FF);
// And let the other char output down below
break;
default:
if (c < INT_SPACE) {
// As per [JACKSON-208], call can now return:
_throwUnquotedSpace(c, "string value");
} else {
// Is this good enough error message?
_reportInvalidChar(c);
}
}
// Need more room?
if (outPtr >= outBuf.length) {
outBuf = _textBuffer.finishCurrentSegment();
outPtr = 0;
}
// Ok, let's add char to output:
outBuf[outPtr++] = (char) c;
}
_textBuffer.setCurrentLength(outPtr);
}
Method called to skim through rest of unparsed String value,
if it is not needed. This can be done bit faster if contents
need not be stored for future access.
/**
* Method called to skim through rest of unparsed String value,
* if it is not needed. This can be done bit faster if contents
* need not be stored for future access.
*/
protected void _skipString() throws IOException
{
_tokenIncomplete = false;
// Need to be fully UTF-8 aware here:
final int[] codes = _icUTF8;
final byte[] inputBuffer = _inputBuffer;
main_loop:
while (true) {
int c;
ascii_loop:
while (true) {
int ptr = _inputPtr;
int max = _inputEnd;
if (ptr >= max) {
_loadMoreGuaranteed();
ptr = _inputPtr;
max = _inputEnd;
}
while (ptr < max) {
c = (int) inputBuffer[ptr++] & 0xFF;
if (codes[c] != 0) {
_inputPtr = ptr;
break ascii_loop;
}
}
_inputPtr = ptr;
}
// Ok: end marker, escape or multi-byte?
if (c == INT_QUOTE) {
break main_loop;
}
switch (codes[c]) {
case 1: // backslash
_decodeEscaped();
break;
case 2: // 2-byte UTF
_skipUtf8_2();
break;
case 3: // 3-byte UTF
_skipUtf8_3();
break;
case 4: // 4-byte UTF
_skipUtf8_4(c);
break;
default:
if (c < INT_SPACE) {
_throwUnquotedSpace(c, "string value");
} else {
// Is this good enough error message?
_reportInvalidChar(c);
}
}
}
}
Method for handling cases where first non-space character
of an expected value token is not legal for standard JSON content.
/**
* Method for handling cases where first non-space character
* of an expected value token is not legal for standard JSON content.
*/
protected JsonToken _handleUnexpectedValue(int c) throws IOException
{
// Most likely an error, unless we are to allow single-quote-strings
switch (c) {
/* This check proceeds only if `Feature.ALLOW_MISSING_VALUES` is enabled;
* it is for missing values. In case of missing values in an array the next token
* will be either ',' or ']'. This case, decrements the already incremented _inputPtr
* in the buffer in case of comma (`,`) so that the existing flow goes back to checking
* the next token which will be comma again and it parsing continues.
* Also the case returns NULL as current token in case of ',' or ']'.
*/
case ']':
if (!_parsingContext.inArray()) {
break;
}
// fall through
case ',':
// 28-Mar-2016: [core#116]: If Feature.ALLOW_MISSING_VALUES is enabled
// we may allow "missing values", that is, encountering a trailing
// comma or closing marker where value would be expected
if ((_features & FEAT_MASK_ALLOW_MISSING) != 0) {
--_inputPtr;
return JsonToken.VALUE_NULL;
}
// fall through
case '}':
// Error: neither is valid at this point; valid closers have
// been handled earlier
_reportUnexpectedChar(c, "expected a value");
case '\'':
if ((_features & FEAT_MASK_ALLOW_SINGLE_QUOTES) != 0) {
return _handleApos();
}
break;
case 'N':
_matchToken("NaN", 1);
if ((_features & FEAT_MASK_NON_NUM_NUMBERS) != 0) {
return resetAsNaN("NaN", Double.NaN);
}
_reportError("Non-standard token 'NaN': enable JsonParser.Feature.ALLOW_NON_NUMERIC_NUMBERS to allow");
break;
case 'I':
_matchToken("Infinity", 1);
if ((_features & FEAT_MASK_NON_NUM_NUMBERS) != 0) {
return resetAsNaN("Infinity", Double.POSITIVE_INFINITY);
}
_reportError("Non-standard token 'Infinity': enable JsonParser.Feature.ALLOW_NON_NUMERIC_NUMBERS to allow");
break;
case '+': // note: '-' is taken as number
if (_inputPtr >= _inputEnd) {
if (!_loadMore()) {
_reportInvalidEOFInValue(JsonToken.VALUE_NUMBER_INT);
}
}
return _handleInvalidNumberStart(_inputBuffer[_inputPtr++] & 0xFF, false);
}
// [core#77] Try to decode most likely token
if (Character.isJavaIdentifierStart(c)) {
_reportInvalidToken(""+((char) c), _validJsonTokenList());
}
// but if it doesn't look like a token:
_reportUnexpectedChar(c, "expected a valid value "+_validJsonValueList());
return null;
}
protected JsonToken _handleApos() throws IOException
{
int c = 0;
// Otherwise almost verbatim copy of _finishString()
int outPtr = 0;
char[] outBuf = _textBuffer.emptyAndGetCurrentSegment();
// Here we do want to do full decoding, hence:
final int[] codes = _icUTF8;
final byte[] inputBuffer = _inputBuffer;
main_loop:
while (true) {
// Then the tight ascii non-funny-char loop:
ascii_loop:
while (true) {
if (_inputPtr >= _inputEnd) {
_loadMoreGuaranteed();
}
if (outPtr >= outBuf.length) {
outBuf = _textBuffer.finishCurrentSegment();
outPtr = 0;
}
int max = _inputEnd;
{
int max2 = _inputPtr + (outBuf.length - outPtr);
if (max2 < max) {
max = max2;
}
}
while (_inputPtr < max) {
c = (int) inputBuffer[_inputPtr++] & 0xFF;
if (c == INT_APOS || codes[c] != 0) {
break ascii_loop;
}
outBuf[outPtr++] = (char) c;
}
}
// Ok: end marker, escape or multi-byte?
if (c == INT_APOS) {
break main_loop;
}
switch (codes[c]) {
case 1: // backslash
c = _decodeEscaped();
break;
case 2: // 2-byte UTF
c = _decodeUtf8_2(c);
break;
case 3: // 3-byte UTF
if ((_inputEnd - _inputPtr) >= 2) {
c = _decodeUtf8_3fast(c);
} else {
c = _decodeUtf8_3(c);
}
break;
case 4: // 4-byte UTF
c = _decodeUtf8_4(c);
// Let's add first part right away:
outBuf[outPtr++] = (char) (0xD800 | (c >> 10));
if (outPtr >= outBuf.length) {
outBuf = _textBuffer.finishCurrentSegment();
outPtr = 0;
}
c = 0xDC00 | (c & 0x3FF);
// And let the other char output down below
break;
default:
if (c < INT_SPACE) {
_throwUnquotedSpace(c, "string value");
}
// Is this good enough error message?
_reportInvalidChar(c);
}
// Need more room?
if (outPtr >= outBuf.length) {
outBuf = _textBuffer.finishCurrentSegment();
outPtr = 0;
}
// Ok, let's add char to output:
outBuf[outPtr++] = (char) c;
}
_textBuffer.setCurrentLength(outPtr);
return JsonToken.VALUE_STRING;
}
/*
/**********************************************************
/* Internal methods, well-known token decoding
/**********************************************************
*/
Method called if expected numeric value (due to leading sign) does not
look like a number
/**
* Method called if expected numeric value (due to leading sign) does not
* look like a number
*/
protected JsonToken _handleInvalidNumberStart(int ch, boolean neg) throws IOException
{
while (ch == 'I') {
if (_inputPtr >= _inputEnd) {
if (!_loadMore()) {
_reportInvalidEOFInValue(JsonToken.VALUE_NUMBER_FLOAT); // possibly?
}
}
ch = _inputBuffer[_inputPtr++];
String match;
if (ch == 'N') {
match = neg ? "-INF" :"+INF";
} else if (ch == 'n') {
match = neg ? "-Infinity" :"+Infinity";
} else {
break;
}
_matchToken(match, 3);
if ((_features & FEAT_MASK_NON_NUM_NUMBERS) != 0) {
return resetAsNaN(match, neg ? Double.NEGATIVE_INFINITY : Double.POSITIVE_INFINITY);
}
_reportError("Non-standard token '%s': enable JsonParser.Feature.ALLOW_NON_NUMERIC_NUMBERS to allow",
match);
}
reportUnexpectedNumberChar(ch, "expected digit (0-9) to follow minus sign, for valid numeric value");
return null;
}
// NOTE: first character already decoded
protected final void _matchTrue() throws IOException
{
int ptr = _inputPtr;
if ((ptr + 3) < _inputEnd) {
byte[] buf = _inputBuffer;
if ((buf[ptr++] == 'r')
&& (buf[ptr++] == 'u')
&& (buf[ptr++] == 'e')) {
int ch = buf[ptr] & 0xFF;
if (ch < INT_0 || (ch == INT_RBRACKET) || (ch == INT_RCURLY)) { // expected/allowed chars
_inputPtr = ptr;
return;
}
}
}
_matchToken2("true", 1);
}
protected final void _matchFalse() throws IOException
{
int ptr = _inputPtr;
if ((ptr + 4) < _inputEnd) {
byte[] buf = _inputBuffer;
if ((buf[ptr++] == 'a')
&& (buf[ptr++] == 'l')
&& (buf[ptr++] == 's')
&& (buf[ptr++] == 'e')) {
int ch = buf[ptr] & 0xFF;
if (ch < INT_0 || (ch == INT_RBRACKET) || (ch == INT_RCURLY)) { // expected/allowed chars
_inputPtr = ptr;
return;
}
}
}
_matchToken2("false", 1);
}
protected final void _matchNull() throws IOException
{
int ptr = _inputPtr;
if ((ptr + 3) < _inputEnd) {
byte[] buf = _inputBuffer;
if ((buf[ptr++] == 'u')
&& (buf[ptr++] == 'l')
&& (buf[ptr++] == 'l')) {
int ch = buf[ptr] & 0xFF;
if (ch < INT_0 || (ch == INT_RBRACKET) || (ch == INT_RCURLY)) { // expected/allowed chars
_inputPtr = ptr;
return;
}
}
}
_matchToken2("null", 1);
}
protected final void _matchToken(String matchStr, int i) throws IOException
{
final int len = matchStr.length();
if ((_inputPtr + len) >= _inputEnd) {
_matchToken2(matchStr, i);
return;
}
do {
if (_inputBuffer[_inputPtr] != matchStr.charAt(i)) {
_reportInvalidToken(matchStr.substring(0, i));
}
++_inputPtr;
} while (++i < len);
int ch = _inputBuffer[_inputPtr] & 0xFF;
if (ch >= '0' && ch != ']' && ch != '}') { // expected/allowed chars
_checkMatchEnd(matchStr, i, ch);
}
}
private final void _matchToken2(String matchStr, int i) throws IOException
{
final int len = matchStr.length();
do {
if (((_inputPtr >= _inputEnd) && !_loadMore())
|| (_inputBuffer[_inputPtr] != matchStr.charAt(i))) {
_reportInvalidToken(matchStr.substring(0, i));
}
++_inputPtr;
} while (++i < len);
// but let's also ensure we either get EOF, or non-alphanum char...
if (_inputPtr >= _inputEnd && !_loadMore()) {
return;
}
int ch = _inputBuffer[_inputPtr] & 0xFF;
if (ch >= '0' && ch != ']' && ch != '}') { // expected/allowed chars
_checkMatchEnd(matchStr, i, ch);
}
}
private final void _checkMatchEnd(String matchStr, int i, int ch) throws IOException {
// but actually only alphanums are problematic
char c = (char) _decodeCharForError(ch);
if (Character.isJavaIdentifierPart(c)) {
_reportInvalidToken(matchStr.substring(0, i));
}
}
/*
/**********************************************************
/* Internal methods, ws skipping, escape/unescape
/**********************************************************
*/
private final int _skipWS() throws IOException
{
while (_inputPtr < _inputEnd) {
int i = _inputBuffer[_inputPtr++] & 0xFF;
if (i > INT_SPACE) {
if (i == INT_SLASH || i == INT_HASH) {
--_inputPtr;
return _skipWS2();
}
return i;
}
if (i != INT_SPACE) {
if (i == INT_LF) {
++_currInputRow;
_currInputRowStart = _inputPtr;
} else if (i == INT_CR) {
_skipCR();
} else if (i != INT_TAB) {
_throwInvalidSpace(i);
}
}
}
return _skipWS2();
}
private final int _skipWS2() throws IOException
{
while (_inputPtr < _inputEnd || _loadMore()) {
int i = _inputBuffer[_inputPtr++] & 0xFF;
if (i > INT_SPACE) {
if (i == INT_SLASH) {
_skipComment();
continue;
}
if (i == INT_HASH) {
if (_skipYAMLComment()) {
continue;
}
}
return i;
}
if (i != INT_SPACE) {
if (i == INT_LF) {
++_currInputRow;
_currInputRowStart = _inputPtr;
} else if (i == INT_CR) {
_skipCR();
} else if (i != INT_TAB) {
_throwInvalidSpace(i);
}
}
}
throw _constructError("Unexpected end-of-input within/between "+_parsingContext.typeDesc()+" entries");
}
private final int _skipWSOrEnd() throws IOException
{
// Let's handle first character separately since it is likely that
// it is either non-whitespace; or we have longer run of white space
if (_inputPtr >= _inputEnd) {
if (!_loadMore()) {
return _eofAsNextChar();
}
}
int i = _inputBuffer[_inputPtr++] & 0xFF;
if (i > INT_SPACE) {
if (i == INT_SLASH || i == INT_HASH) {
--_inputPtr;
return _skipWSOrEnd2();
}
return i;
}
if (i != INT_SPACE) {
if (i == INT_LF) {
++_currInputRow;
_currInputRowStart = _inputPtr;
} else if (i == INT_CR) {
_skipCR();
} else if (i != INT_TAB) {
_throwInvalidSpace(i);
}
}
while (_inputPtr < _inputEnd) {
i = _inputBuffer[_inputPtr++] & 0xFF;
if (i > INT_SPACE) {
if (i == INT_SLASH || i == INT_HASH) {
--_inputPtr;
return _skipWSOrEnd2();
}
return i;
}
if (i != INT_SPACE) {
if (i == INT_LF) {
++_currInputRow;
_currInputRowStart = _inputPtr;
} else if (i == INT_CR) {
_skipCR();
} else if (i != INT_TAB) {
_throwInvalidSpace(i);
}
}
}
return _skipWSOrEnd2();
}
private final int _skipWSOrEnd2() throws IOException
{
while ((_inputPtr < _inputEnd) || _loadMore()) {
int i = _inputBuffer[_inputPtr++] & 0xFF;
if (i > INT_SPACE) {
if (i == INT_SLASH) {
_skipComment();
continue;
}
if (i == INT_HASH) {
if (_skipYAMLComment()) {
continue;
}
}
return i;
} else if (i != INT_SPACE) {
if (i == INT_LF) {
++_currInputRow;
_currInputRowStart = _inputPtr;
} else if (i == INT_CR) {
_skipCR();
} else if (i != INT_TAB) {
_throwInvalidSpace(i);
}
}
}
// We ran out of input...
return _eofAsNextChar();
}
private final int _skipColon() throws IOException
{
if ((_inputPtr + 4) >= _inputEnd) {
return _skipColon2(false);
}
// Fast path: colon with optional single-space/tab before and/or after:
int i = _inputBuffer[_inputPtr];
if (i == INT_COLON) { // common case, no leading space
i = _inputBuffer[++_inputPtr];
if (i > INT_SPACE) { // nor trailing
if (i == INT_SLASH || i == INT_HASH) {
return _skipColon2(true);
}
++_inputPtr;
return i;
}
if (i == INT_SPACE || i == INT_TAB) {
i = (int) _inputBuffer[++_inputPtr];
if (i > INT_SPACE) {
if (i == INT_SLASH || i == INT_HASH) {
return _skipColon2(true);
}
++_inputPtr;
return i;
}
}
return _skipColon2(true); // true -> skipped colon
}
if (i == INT_SPACE || i == INT_TAB) {
i = _inputBuffer[++_inputPtr];
}
if (i == INT_COLON) {
i = _inputBuffer[++_inputPtr];
if (i > INT_SPACE) {
if (i == INT_SLASH || i == INT_HASH) {
return _skipColon2(true);
}
++_inputPtr;
return i;
}
if (i == INT_SPACE || i == INT_TAB) {
i = (int) _inputBuffer[++_inputPtr];
if (i > INT_SPACE) {
if (i == INT_SLASH || i == INT_HASH) {
return _skipColon2(true);
}
++_inputPtr;
return i;
}
}
return _skipColon2(true);
}
return _skipColon2(false);
}
private final int _skipColon2(boolean gotColon) throws IOException
{
while (_inputPtr < _inputEnd || _loadMore()) {
int i = _inputBuffer[_inputPtr++] & 0xFF;
if (i > INT_SPACE) {
if (i == INT_SLASH) {
_skipComment();
continue;
}
if (i == INT_HASH) {
if (_skipYAMLComment()) {
continue;
}
}
if (gotColon) {
return i;
}
if (i != INT_COLON) {
_reportUnexpectedChar(i, "was expecting a colon to separate field name and value");
}
gotColon = true;
} else if (i != INT_SPACE) {
if (i == INT_LF) {
++_currInputRow;
_currInputRowStart = _inputPtr;
} else if (i == INT_CR) {
_skipCR();
} else if (i != INT_TAB) {
_throwInvalidSpace(i);
}
}
}
_reportInvalidEOF(" within/between "+_parsingContext.typeDesc()+" entries",
null);
return -1;
}
private final void _skipComment() throws IOException
{
if ((_features & FEAT_MASK_ALLOW_JAVA_COMMENTS) == 0) {
_reportUnexpectedChar('/', "maybe a (non-standard) comment? (not recognized as one since Feature 'ALLOW_COMMENTS' not enabled for parser)");
}
// First: check which comment (if either) it is:
if (_inputPtr >= _inputEnd && !_loadMore()) {
_reportInvalidEOF(" in a comment", null);
}
int c = _inputBuffer[_inputPtr++] & 0xFF;
if (c == INT_SLASH) {
_skipLine();
} else if (c == INT_ASTERISK) {
_skipCComment();
} else {
_reportUnexpectedChar(c, "was expecting either '*' or '/' for a comment");
}
}
private final void _skipCComment() throws IOException
{
// Need to be UTF-8 aware here to decode content (for skipping)
final int[] codes = CharTypes.getInputCodeComment();
// Ok: need the matching '*/'
main_loop:
while ((_inputPtr < _inputEnd) || _loadMore()) {
int i = (int) _inputBuffer[_inputPtr++] & 0xFF;
int code = codes[i];
if (code != 0) {
switch (code) {
case '*':
if (_inputPtr >= _inputEnd && !_loadMore()) {
break main_loop;
}
if (_inputBuffer[_inputPtr] == INT_SLASH) {
++_inputPtr;
return;
}
break;
case INT_LF:
++_currInputRow;
_currInputRowStart = _inputPtr;
break;
case INT_CR:
_skipCR();
break;
case 2: // 2-byte UTF
_skipUtf8_2();
break;
case 3: // 3-byte UTF
_skipUtf8_3();
break;
case 4: // 4-byte UTF
_skipUtf8_4(i);
break;
default: // e.g. -1
// Is this good enough error message?
_reportInvalidChar(i);
}
}
}
_reportInvalidEOF(" in a comment", null);
}
private final boolean _skipYAMLComment() throws IOException
{
if ((_features & FEAT_MASK_ALLOW_YAML_COMMENTS) == 0) {
return false;
}
_skipLine();
return true;
}
Method for skipping contents of an input line; usually for CPP
and YAML style comments.
/**
* Method for skipping contents of an input line; usually for CPP
* and YAML style comments.
*/
private final void _skipLine() throws IOException
{
// Ok: need to find EOF or linefeed
final int[] codes = CharTypes.getInputCodeComment();
while ((_inputPtr < _inputEnd) || _loadMore()) {
int i = (int) _inputBuffer[_inputPtr++] & 0xFF;
int code = codes[i];
if (code != 0) {
switch (code) {
case INT_LF:
++_currInputRow;
_currInputRowStart = _inputPtr;
return;
case INT_CR:
_skipCR();
return;
case '*': // nop for these comments
break;
case 2: // 2-byte UTF
_skipUtf8_2();
break;
case 3: // 3-byte UTF
_skipUtf8_3();
break;
case 4: // 4-byte UTF
_skipUtf8_4(i);
break;
default: // e.g. -1
if (code < 0) {
// Is this good enough error message?
_reportInvalidChar(i);
}
}
}
}
}
@Override
protected char _decodeEscaped() throws IOException
{
if (_inputPtr >= _inputEnd) {
if (!_loadMore()) {
_reportInvalidEOF(" in character escape sequence", JsonToken.VALUE_STRING);
}
}
int c = (int) _inputBuffer[_inputPtr++];
switch (c) {
// First, ones that are mapped
case 'b':
return '\b';
case 't':
return '\t';
case 'n':
return '\n';
case 'f':
return '\f';
case 'r':
return '\r';
// And these are to be returned as they are
case '"':
case '/':
case '\\':
return (char) c;
case 'u': // and finally hex-escaped
break;
default:
return _handleUnrecognizedCharacterEscape((char) _decodeCharForError(c));
}
// Ok, a hex escape. Need 4 characters
int value = 0;
for (int i = 0; i < 4; ++i) {
if (_inputPtr >= _inputEnd) {
if (!_loadMore()) {
_reportInvalidEOF(" in character escape sequence", JsonToken.VALUE_STRING);
}
}
int ch = _inputBuffer[_inputPtr++];
int digit = CharTypes.charToHex(ch);
if (digit < 0) {
_reportUnexpectedChar(ch & 0xFF, "expected a hex-digit for character escape sequence");
}
value = (value << 4) | digit;
}
return (char) value;
}
protected int _decodeCharForError(int firstByte) throws IOException
{
int c = firstByte & 0xFF;
if (c > 0x7F) { // if >= 0, is ascii and fine as is
int needed;
// Ok; if we end here, we got multi-byte combination
if ((c & 0xE0) == 0xC0) { // 2 bytes (0x0080 - 0x07FF)
c &= 0x1F;
needed = 1;
} else if ((c & 0xF0) == 0xE0) { // 3 bytes (0x0800 - 0xFFFF)
c &= 0x0F;
needed = 2;
} else if ((c & 0xF8) == 0xF0) {
// 4 bytes; double-char with surrogates and all...
c &= 0x07;
needed = 3;
} else {
_reportInvalidInitial(c & 0xFF);
needed = 1; // never gets here
}
int d = nextByte();
if ((d & 0xC0) != 0x080) {
_reportInvalidOther(d & 0xFF);
}
c = (c << 6) | (d & 0x3F);
if (needed > 1) { // needed == 1 means 2 bytes total
d = nextByte(); // 3rd byte
if ((d & 0xC0) != 0x080) {
_reportInvalidOther(d & 0xFF);
}
c = (c << 6) | (d & 0x3F);
if (needed > 2) { // 4 bytes? (need surrogates)
d = nextByte();
if ((d & 0xC0) != 0x080) {
_reportInvalidOther(d & 0xFF);
}
c = (c << 6) | (d & 0x3F);
}
}
}
return c;
}
/*
/**********************************************************
/* Internal methods,UTF8 decoding
/**********************************************************
*/
private final int _decodeUtf8_2(int c) throws IOException
{
if (_inputPtr >= _inputEnd) {
_loadMoreGuaranteed();
}
int d = (int) _inputBuffer[_inputPtr++];
if ((d & 0xC0) != 0x080) {
_reportInvalidOther(d & 0xFF, _inputPtr);
}
return ((c & 0x1F) << 6) | (d & 0x3F);
}
private final int _decodeUtf8_3(int c1) throws IOException
{
if (_inputPtr >= _inputEnd) {
_loadMoreGuaranteed();
}
c1 &= 0x0F;
int d = (int) _inputBuffer[_inputPtr++];
if ((d & 0xC0) != 0x080) {
_reportInvalidOther(d & 0xFF, _inputPtr);
}
int c = (c1 << 6) | (d & 0x3F);
if (_inputPtr >= _inputEnd) {
_loadMoreGuaranteed();
}
d = (int) _inputBuffer[_inputPtr++];
if ((d & 0xC0) != 0x080) {
_reportInvalidOther(d & 0xFF, _inputPtr);
}
c = (c << 6) | (d & 0x3F);
return c;
}
private final int _decodeUtf8_3fast(int c1) throws IOException
{
c1 &= 0x0F;
int d = (int) _inputBuffer[_inputPtr++];
if ((d & 0xC0) != 0x080) {
_reportInvalidOther(d & 0xFF, _inputPtr);
}
int c = (c1 << 6) | (d & 0x3F);
d = (int) _inputBuffer[_inputPtr++];
if ((d & 0xC0) != 0x080) {
_reportInvalidOther(d & 0xFF, _inputPtr);
}
c = (c << 6) | (d & 0x3F);
return c;
}
Returns: Character value minus 0x10000; this so that caller
can readily expand it to actual surrogates
/**
* @return Character value <b>minus 0x10000</c>; this so that caller
* can readily expand it to actual surrogates
*/
private final int _decodeUtf8_4(int c) throws IOException
{
if (_inputPtr >= _inputEnd) {
_loadMoreGuaranteed();
}
int d = (int) _inputBuffer[_inputPtr++];
if ((d & 0xC0) != 0x080) {
_reportInvalidOther(d & 0xFF, _inputPtr);
}
c = ((c & 0x07) << 6) | (d & 0x3F);
if (_inputPtr >= _inputEnd) {
_loadMoreGuaranteed();
}
d = (int) _inputBuffer[_inputPtr++];
if ((d & 0xC0) != 0x080) {
_reportInvalidOther(d & 0xFF, _inputPtr);
}
c = (c << 6) | (d & 0x3F);
if (_inputPtr >= _inputEnd) {
_loadMoreGuaranteed();
}
d = (int) _inputBuffer[_inputPtr++];
if ((d & 0xC0) != 0x080) {
_reportInvalidOther(d & 0xFF, _inputPtr);
}
/* note: won't change it to negative here, since caller
* already knows it'll need a surrogate
*/
return ((c << 6) | (d & 0x3F)) - 0x10000;
}
private final void _skipUtf8_2() throws IOException
{
if (_inputPtr >= _inputEnd) {
_loadMoreGuaranteed();
}
int c = (int) _inputBuffer[_inputPtr++];
if ((c & 0xC0) != 0x080) {
_reportInvalidOther(c & 0xFF, _inputPtr);
}
}
/* Alas, can't heavily optimize skipping, since we still have to
* do validity checks...
*/
private final void _skipUtf8_3() throws IOException
{
if (_inputPtr >= _inputEnd) {
_loadMoreGuaranteed();
}
//c &= 0x0F;
int c = (int) _inputBuffer[_inputPtr++];
if ((c & 0xC0) != 0x080) {
_reportInvalidOther(c & 0xFF, _inputPtr);
}
if (_inputPtr >= _inputEnd) {
_loadMoreGuaranteed();
}
c = (int) _inputBuffer[_inputPtr++];
if ((c & 0xC0) != 0x080) {
_reportInvalidOther(c & 0xFF, _inputPtr);
}
}
private final void _skipUtf8_4(int c) throws IOException
{
if (_inputPtr >= _inputEnd) {
_loadMoreGuaranteed();
}
int d = (int) _inputBuffer[_inputPtr++];
if ((d & 0xC0) != 0x080) {
_reportInvalidOther(d & 0xFF, _inputPtr);
}
if (_inputPtr >= _inputEnd) {
_loadMoreGuaranteed();
}
d = (int) _inputBuffer[_inputPtr++];
if ((d & 0xC0) != 0x080) {
_reportInvalidOther(d & 0xFF, _inputPtr);
}
if (_inputPtr >= _inputEnd) {
_loadMoreGuaranteed();
}
d = (int) _inputBuffer[_inputPtr++];
if ((d & 0xC0) != 0x080) {
_reportInvalidOther(d & 0xFF, _inputPtr);
}
}
/*
/**********************************************************
/* Internal methods, input loading
/**********************************************************
*/
We actually need to check the character value here
(to see if we have \n following \r).
/**
* We actually need to check the character value here
* (to see if we have \n following \r).
*/
protected final void _skipCR() throws IOException
{
if (_inputPtr < _inputEnd || _loadMore()) {
if (_inputBuffer[_inputPtr] == BYTE_LF) {
++_inputPtr;
}
}
++_currInputRow;
_currInputRowStart = _inputPtr;
}
private int nextByte() throws IOException
{
if (_inputPtr >= _inputEnd) {
_loadMoreGuaranteed();
}
return _inputBuffer[_inputPtr++] & 0xFF;
}
/*
/**********************************************************
/* Internal methods, error reporting
/**********************************************************
*/
protected void _reportInvalidToken(String matchedPart, int ptr) throws IOException {
_inputPtr = ptr;
_reportInvalidToken(matchedPart, _validJsonTokenList());
}
protected void _reportInvalidToken(String matchedPart) throws IOException {
_reportInvalidToken(matchedPart, _validJsonTokenList());
}
protected void _reportInvalidToken(String matchedPart, String msg) throws IOException
{
/* Let's just try to find what appears to be the token, using
* regular Java identifier character rules. It's just a heuristic,
* nothing fancy here (nor fast).
*/
StringBuilder sb = new StringBuilder(matchedPart);
while ((_inputPtr < _inputEnd) || _loadMore()) {
int i = (int) _inputBuffer[_inputPtr++];
char c = (char) _decodeCharForError(i);
if (!Character.isJavaIdentifierPart(c)) {
// 11-Jan-2016, tatu: note: we will fully consume the character,
// included or not, so if recovery was possible, it'd be off-by-one...
break;
}
sb.append(c);
if (sb.length() >= MAX_ERROR_TOKEN_LENGTH) {
sb.append("...");
break;
}
}
_reportError("Unrecognized token '%s': was expecting %s", sb, msg);
}
protected void _reportInvalidChar(int c) throws JsonParseException
{
// Either invalid WS or illegal UTF-8 start char
if (c < INT_SPACE) {
_throwInvalidSpace(c);
}
_reportInvalidInitial(c);
}
protected void _reportInvalidInitial(int mask) throws JsonParseException {
_reportError("Invalid UTF-8 start byte 0x"+Integer.toHexString(mask));
}
protected void _reportInvalidOther(int mask) throws JsonParseException {
_reportError("Invalid UTF-8 middle byte 0x"+Integer.toHexString(mask));
}
protected void _reportInvalidOther(int mask, int ptr)
throws JsonParseException
{
_inputPtr = ptr;
_reportInvalidOther(mask);
}
/*
/**********************************************************
/* Internal methods, binary access
/**********************************************************
*/
Efficient handling for incremental parsing of base64-encoded
textual content.
/**
* Efficient handling for incremental parsing of base64-encoded
* textual content.
*/
@SuppressWarnings("resource")
protected final byte[] _decodeBase64(Base64Variant b64variant) throws IOException
{
ByteArrayBuilder builder = _getByteArrayBuilder();
while (true) {
// first, we'll skip preceding white space, if any
int ch;
do {
if (_inputPtr >= _inputEnd) {
_loadMoreGuaranteed();
}
ch = (int) _inputBuffer[_inputPtr++] & 0xFF;
} while (ch <= INT_SPACE);
int bits = b64variant.decodeBase64Char(ch);
if (bits < 0) { // reached the end, fair and square?
if (ch == INT_QUOTE) {
return builder.toByteArray();
}
bits = _decodeBase64Escape(b64variant, ch, 0);
if (bits < 0) { // white space to skip
continue;
}
}
int decodedData = bits;
// then second base64 char; can't get padding yet, nor ws
if (_inputPtr >= _inputEnd) {
_loadMoreGuaranteed();
}
ch = _inputBuffer[_inputPtr++] & 0xFF;
bits = b64variant.decodeBase64Char(ch);
if (bits < 0) {
bits = _decodeBase64Escape(b64variant, ch, 1);
}
decodedData = (decodedData << 6) | bits;
// third base64 char; can be padding, but not ws
if (_inputPtr >= _inputEnd) {
_loadMoreGuaranteed();
}
ch = _inputBuffer[_inputPtr++] & 0xFF;
bits = b64variant.decodeBase64Char(ch);
// First branch: can get padding (-> 1 byte)
if (bits < 0) {
if (bits != Base64Variant.BASE64_VALUE_PADDING) {
// could also just be 'missing' padding
if (ch == INT_QUOTE) {
decodedData >>= 4;
builder.append(decodedData);
if (b64variant.usesPadding()) {
--_inputPtr; // to keep parser state bit more consistent
_handleBase64MissingPadding(b64variant);
}
return builder.toByteArray();
}
bits = _decodeBase64Escape(b64variant, ch, 2);
}
if (bits == Base64Variant.BASE64_VALUE_PADDING) {
// Ok, must get padding
if (_inputPtr >= _inputEnd) {
_loadMoreGuaranteed();
}
ch = _inputBuffer[_inputPtr++] & 0xFF;
if (!b64variant.usesPaddingChar(ch)) {
if (_decodeBase64Escape(b64variant, ch, 3) != Base64Variant.BASE64_VALUE_PADDING) {
throw reportInvalidBase64Char(b64variant, ch, 3, "expected padding character '"+b64variant.getPaddingChar()+"'");
}
}
// Got 12 bits, only need 8, need to shift
decodedData >>= 4;
builder.append(decodedData);
continue;
}
}
// Nope, 2 or 3 bytes
decodedData = (decodedData << 6) | bits;
// fourth and last base64 char; can be padding, but not ws
if (_inputPtr >= _inputEnd) {
_loadMoreGuaranteed();
}
ch = _inputBuffer[_inputPtr++] & 0xFF;
bits = b64variant.decodeBase64Char(ch);
if (bits < 0) {
if (bits != Base64Variant.BASE64_VALUE_PADDING) {
// could also just be 'missing' padding
if (ch == INT_QUOTE) {
decodedData >>= 2;
builder.appendTwoBytes(decodedData);
if (b64variant.usesPadding()) {
--_inputPtr; // to keep parser state bit more consistent
_handleBase64MissingPadding(b64variant);
}
return builder.toByteArray();
}
bits = _decodeBase64Escape(b64variant, ch, 3);
}
if (bits == Base64Variant.BASE64_VALUE_PADDING) {
// With padding we only get 2 bytes; but we have to shift it
// a bit so it is identical to triplet case with partial output.
// 3 chars gives 3x6 == 18 bits, of which 2 are dummies, need to discard:
decodedData >>= 2;
builder.appendTwoBytes(decodedData);
continue;
}
}
// otherwise, our triplet is now complete
decodedData = (decodedData << 6) | bits;
builder.appendThreeBytes(decodedData);
}
}
/*
/**********************************************************
/* Improved location updating (refactored in 2.7)
/**********************************************************
*/
// As per [core#108], must ensure we call the right method
@Override
public JsonLocation getTokenLocation()
{
if (_currToken == JsonToken.FIELD_NAME) {
long total = _currInputProcessed + (_nameStartOffset-1);
return new JsonLocation(_getSourceReference(),
total, -1L, _nameStartRow, _nameStartCol);
}
return new JsonLocation(_getSourceReference(),
_tokenInputTotal-1, -1L, _tokenInputRow, _tokenInputCol);
}
// As per [core#108], must ensure we call the right method
@Override
public JsonLocation getCurrentLocation()
{
int col = _inputPtr - _currInputRowStart + 1; // 1-based
return new JsonLocation(_getSourceReference(),
_currInputProcessed + _inputPtr, -1L, // bytes, chars
_currInputRow, col);
}
// @since 2.7
private final void _updateLocation()
{
_tokenInputRow = _currInputRow;
final int ptr = _inputPtr;
_tokenInputTotal = _currInputProcessed + ptr;
_tokenInputCol = ptr - _currInputRowStart;
}
// @since 2.7
private final void _updateNameLocation()
{
_nameStartRow = _currInputRow;
final int ptr = _inputPtr;
_nameStartOffset = ptr;
_nameStartCol = ptr - _currInputRowStart;
}
/*
/**********************************************************
/* Internal methods, other
/**********************************************************
*/
private final JsonToken _closeScope(int i) throws JsonParseException {
if (i == INT_RCURLY) {
_closeObjectScope();
return (_currToken = JsonToken.END_OBJECT);
}
_closeArrayScope();
return (_currToken = JsonToken.END_ARRAY);
}
private final void _closeArrayScope() throws JsonParseException {
_updateLocation();
if (!_parsingContext.inArray()) {
_reportMismatchedEndMarker(']', '}');
}
_parsingContext = _parsingContext.clearAndGetParent();
}
private final void _closeObjectScope() throws JsonParseException {
_updateLocation();
if (!_parsingContext.inObject()) {
_reportMismatchedEndMarker('}', ']');
}
_parsingContext = _parsingContext.clearAndGetParent();
}
}