Index: experimental/PdfViewer/pdfparser/native/SkPdfNativeTokenizer.h |
=================================================================== |
--- experimental/PdfViewer/pdfparser/native/SkPdfNativeTokenizer.h (revision 9879) |
+++ experimental/PdfViewer/pdfparser/native/SkPdfNativeTokenizer.h (working copy) |
@@ -1,10 +1,145 @@ |
#ifndef EXPERIMENTAL_PDFVIEWER_PDFPARSER_NATIVE_SKPDFNATIVETOKENIZER_H_ |
#define EXPERIMENTAL_PDFVIEWER_PDFPARSER_NATIVE_SKPDFNATIVETOKENIZER_H_ |
-class SkPdfNativeTokenizer : public SkPdfPodofoTokenizer { |
+#include "SkTDArray.h" |
+#include "SkTDict.h" |
+#include <math.h> |
+#include <string.h> |
+ |
+class SkPdfMapper; |
+class SkPdfDictionary; |
+ |
+// White Spaces |
+#define kNUL_PdfWhiteSpace '\x00' |
+#define kHT_PdfWhiteSpace '\x09' |
+#define kLF_PdfWhiteSpace '\x0A' |
+#define kFF_PdfWhiteSpace '\x0C' |
+#define kCR_PdfWhiteSpace '\x0D' |
+#define kSP_PdfWhiteSpace '\x20' |
+ |
+// PdfDelimiters |
+#define kOpenedRoundBracket_PdfDelimiter '(' |
+#define kClosedRoundBracket_PdfDelimiter ')' |
+#define kOpenedInequityBracket_PdfDelimiter '<' |
+#define kClosedInequityBracket_PdfDelimiter '>' |
+#define kOpenedSquareBracket_PdfDelimiter '[' |
+#define kClosedSquareBracket_PdfDelimiter ']' |
+#define kOpenedCurlyBracket_PdfDelimiter '{' |
+#define kClosedCurlyBracket_PdfDelimiter '}' |
+#define kNamed_PdfDelimiter '/' |
+#define kComment_PdfDelimiter '%' |
+ |
+#define kEscape_PdfSpecial '\\' |
+#define kBackspace_PdfSpecial '\x08' |
+ |
+// TODO(edisonn): what is the faster way for compiler/machine type to evaluate this expressions? |
+// we should evaluate all options. might be even different from one machine to another |
+// 1) expand expression, let compiler optimize it |
+// 2) binary search |
+// 3) linear search in array |
+// 4) vector (e.f. T type[256] .. return type[ch] ... |
+// 5) manually build the expression with least number of operators, e.g. for consecutive |
+// chars, we can use an binary equal ignoring last bit |
+#define isPdfWhiteSpace(ch) (((ch)==kNUL_PdfWhiteSpace)||((ch)==kHT_PdfWhiteSpace)||((ch)==kLF_PdfWhiteSpace)||((ch)==kFF_PdfWhiteSpace)||((ch)==kCR_PdfWhiteSpace)||((ch)==kSP_PdfWhiteSpace)) |
+ |
+#define isPdfEOL(ch) (((ch)==kLF_PdfWhiteSpace)||((ch)==kCR_PdfWhiteSpace)) |
+ |
+ |
+#define isPdfDelimiter(ch) (((ch)==kOpenedRoundBracket_PdfDelimiter)||\ |
+ ((ch)==kClosedRoundBracket_PdfDelimiter)||\ |
+ ((ch)==kOpenedInequityBracket_PdfDelimiter)||\ |
+ ((ch)==kClosedInequityBracket_PdfDelimiter)||\ |
+ ((ch)==kOpenedSquareBracket_PdfDelimiter)||\ |
+ ((ch)==kClosedSquareBracket_PdfDelimiter)||\ |
+ ((ch)==kOpenedCurlyBracket_PdfDelimiter)||\ |
+ ((ch)==kClosedCurlyBracket_PdfDelimiter)||\ |
+ ((ch)==kNamed_PdfDelimiter)||\ |
+ ((ch)==kComment_PdfDelimiter)) |
+ |
+#define isPdfWhiteSpaceOrPdfDelimiter(ch) (isPdfWhiteSpace(ch)||isPdfDelimiter(ch)) |
+ |
+#define isPdfDigit(ch) ((ch)>='0'&&(ch)<='9') |
+#define isPdfNumeric(ch) (isPdfDigit(ch)||(ch)=='+'||(ch)=='-') |
+ |
+unsigned char* skipPdfWhiteSpaces(unsigned char* buffer, size_t len); |
+unsigned char* endOfPdfToken(unsigned char* start, size_t len); |
+unsigned char* skipPdfComment(unsigned char* start, size_t len); |
+ |
+// TODO(edisonn): typedef read and integer tyepes? make less readable... |
+//typedef double SkPdfReal; |
+//typedef int64_t SkPdfInteger; |
+ |
+// an allocator only allocates memory, and it deletes it all when the allocator is destroyed |
+// this would allow us not to do any garbage collection while we parse or draw a pdf, and defere it |
+// while the user is looking at the image |
+ |
+class SkPdfObject; |
+ |
+class SkPdfAllocator { |
+#define BUFFER_SIZE 1024 |
+ SkTDArray<SkPdfObject*> fHistory; |
+ SkTDArray<void*> fHandles; |
+ SkPdfObject* fCurrent; |
+ int fCurrentUsed; |
+ |
+ SkPdfObject* allocBlock(); |
+ |
public: |
- SkPdfNativeTokenizer(); |
+ SkPdfAllocator() { |
+ fCurrent = allocBlock(); |
+ fCurrentUsed = 0; |
+ } |
+ |
+ ~SkPdfAllocator(); |
+ |
+ SkPdfObject* allocObject(); |
+ |
+ // TODO(edisonn): free this memory in destructor, track the usage? |
+ void* alloc(size_t bytes) { |
+ void* data = malloc(bytes); |
+ fHandles.push(data); |
+ return data; |
+ } |
+}; |
+ |
+unsigned char* nextObject(unsigned char* start, unsigned char* end, SkPdfObject* token, SkPdfAllocator* allocator); |
+ |
+enum SkPdfTokenType { |
+ kKeyword_TokenType, |
+ kObject_TokenType, |
+}; |
+ |
+struct PdfToken { |
+ const char* fKeyword; |
+ size_t fKeywordLength; |
+ SkPdfObject* fObject; |
+ SkPdfTokenType fType; |
+ |
+ PdfToken() : fKeyword(NULL), fKeywordLength(0), fObject(NULL) {} |
+}; |
+ |
+class SkPdfNativeTokenizer { |
+public: |
+ SkPdfNativeTokenizer(SkPdfObject* objWithStream, const SkPdfMapper* mapper, SkPdfAllocator* allocator); |
+ SkPdfNativeTokenizer(unsigned char* buffer, int len, const SkPdfMapper* mapper, SkPdfAllocator* allocator); |
+ |
virtual ~SkPdfNativeTokenizer(); |
+ |
+ bool readToken(PdfToken* token); |
+ bool readTokenCore(PdfToken* token); |
+ void PutBack(PdfToken token); |
+ |
+private: |
+ const SkPdfMapper* fMapper; |
+ SkPdfAllocator* fAllocator; |
+ |
+ unsigned char* fUncompressedStreamStart; |
+ unsigned char* fUncompressedStream; |
+ unsigned char* fUncompressedStreamEnd; |
+ |
+ bool fEmpty; |
+ bool fHasPutBack; |
+ PdfToken fPutBack; |
}; |
#endif // EXPERIMENTAL_PDFVIEWER_PDFPARSER_NATIVE_SKPDFNATIVETOKENIZER_H_ |