Index: xfa/src/fxfa/fm2js/xfa_lexer.cpp |
diff --git a/xfa/src/fxfa/fm2js/xfa_lexer.cpp b/xfa/src/fxfa/fm2js/xfa_lexer.cpp |
deleted file mode 100644 |
index f93fa839d37a78629f2834f89f71790e48b550cf..0000000000000000000000000000000000000000 |
--- a/xfa/src/fxfa/fm2js/xfa_lexer.cpp |
+++ /dev/null |
@@ -1,552 +0,0 @@ |
-// Copyright 2014 PDFium Authors. All rights reserved. |
-// Use of this source code is governed by a BSD-style license that can be |
-// found in the LICENSE file. |
- |
-// Original code copyright 2014 Foxit Software Inc. http://www.foxitsoftware.com |
- |
-#include "xfa/src/fxfa/fm2js/xfa_lexer.h" |
- |
-#include "core/include/fxcrt/fx_ext.h" |
- |
-namespace { |
- |
-struct XFA_FMDChar { |
- static const FX_WCHAR* inc(const FX_WCHAR*& p) { |
- ++p; |
- return p; |
- } |
- static const FX_WCHAR* dec(const FX_WCHAR*& p) { |
- --p; |
- return p; |
- } |
- static uint16_t get(const FX_WCHAR* p) { return *p; } |
- static FX_BOOL isWhiteSpace(const FX_WCHAR* p) { |
- return (*p) == 0x09 || (*p) == 0x0b || (*p) == 0x0c || (*p) == 0x20; |
- } |
- static FX_BOOL isLineTerminator(const FX_WCHAR* p) { |
- return *p == 0x0A || *p == 0x0D; |
- } |
- static FX_BOOL isBinary(const FX_WCHAR* p) { |
- return (*p) >= '0' && (*p) <= '1'; |
- } |
- static FX_BOOL isOctal(const FX_WCHAR* p) { |
- return (*p) >= '0' && (*p) <= '7'; |
- } |
- static FX_BOOL isDigital(const FX_WCHAR* p) { |
- return (*p) >= '0' && (*p) <= '9'; |
- } |
- static FX_BOOL isHex(const FX_WCHAR* p) { |
- return isDigital(p) || ((*p) >= 'a' && (*p) <= 'f') || |
- ((*p) >= 'A' && (*p) <= 'F'); |
- } |
- static FX_BOOL isAlpha(const FX_WCHAR* p) { |
- return ((*p) <= 'z' && (*p) >= 'a') || ((*p) <= 'Z' && (*p) >= 'A'); |
- } |
- static FX_BOOL isAvalid(const FX_WCHAR* p, FX_BOOL flag = 0); |
- static FX_BOOL string2number(const FX_WCHAR* s, |
- FX_DOUBLE* pValue, |
- const FX_WCHAR*& pEnd); |
- static FX_BOOL isUnicodeAlpha(uint16_t ch); |
-}; |
- |
-inline FX_BOOL XFA_FMDChar::isAvalid(const FX_WCHAR* p, FX_BOOL flag) { |
- if (*p == 0) { |
- return 1; |
- } |
- if ((*p <= 0x0A && *p >= 0x09) || *p == 0x0D || |
- (*p <= 0xd7ff && *p >= 0x20) || (*p <= 0xfffd && *p >= 0xe000)) { |
- return 1; |
- } |
- if (!flag) { |
- if (*p == 0x0B || *p == 0x0C) { |
- return 1; |
- } |
- } |
- return 0; |
-} |
- |
-inline FX_BOOL XFA_FMDChar::string2number(const FX_WCHAR* s, |
- FX_DOUBLE* pValue, |
- const FX_WCHAR*& pEnd) { |
- if (s) { |
- *pValue = wcstod((wchar_t*)s, (wchar_t**)&pEnd); |
- } |
- return 0; |
-} |
- |
-inline FX_BOOL XFA_FMDChar::isUnicodeAlpha(uint16_t ch) { |
- if (ch == 0 || ch == 0x0A || ch == 0x0D || ch == 0x09 || ch == 0x0B || |
- ch == 0x0C || ch == 0x20 || ch == '.' || ch == ';' || ch == '"' || |
- ch == '=' || ch == '<' || ch == '>' || ch == ',' || ch == '(' || |
- ch == ')' || ch == ']' || ch == '[' || ch == '&' || ch == '|' || |
- ch == '+' || ch == '-' || ch == '*' || ch == '/') { |
- return FALSE; |
- } |
- return TRUE; |
-} |
- |
-const XFA_FMKeyword keyWords[] = { |
- {TOKand, 0x00000026, L"&"}, |
- {TOKlparen, 0x00000028, L"("}, |
- {TOKrparen, 0x00000029, L")"}, |
- {TOKmul, 0x0000002a, L"*"}, |
- {TOKplus, 0x0000002b, L"+"}, |
- {TOKcomma, 0x0000002c, L","}, |
- {TOKminus, 0x0000002d, L"-"}, |
- {TOKdot, 0x0000002e, L"."}, |
- {TOKdiv, 0x0000002f, L"/"}, |
- {TOKlt, 0x0000003c, L"<"}, |
- {TOKassign, 0x0000003d, L"="}, |
- {TOKgt, 0x0000003e, L">"}, |
- {TOKlbracket, 0x0000005b, L"["}, |
- {TOKrbracket, 0x0000005d, L"]"}, |
- {TOKor, 0x0000007c, L"|"}, |
- {TOKdotscream, 0x0000ec11, L".#"}, |
- {TOKdotstar, 0x0000ec18, L".*"}, |
- {TOKdotdot, 0x0000ec1c, L".."}, |
- {TOKle, 0x000133f9, L"<="}, |
- {TOKne, 0x000133fa, L"<>"}, |
- {TOKeq, 0x0001391a, L"=="}, |
- {TOKge, 0x00013e3b, L">="}, |
- {TOKdo, 0x00020153, L"do"}, |
- {TOKkseq, 0x00020676, L"eq"}, |
- {TOKksge, 0x000210ac, L"ge"}, |
- {TOKksgt, 0x000210bb, L"gt"}, |
- {TOKif, 0x00021aef, L"if"}, |
- {TOKin, 0x00021af7, L"in"}, |
- {TOKksle, 0x00022a51, L"le"}, |
- {TOKkslt, 0x00022a60, L"lt"}, |
- {TOKksne, 0x00023493, L"ne"}, |
- {TOKksor, 0x000239c1, L"or"}, |
- {TOKnull, 0x052931bb, L"null"}, |
- {TOKbreak, 0x05518c25, L"break"}, |
- {TOKksand, 0x09f9db33, L"and"}, |
- {TOKend, 0x0a631437, L"end"}, |
- {TOKeof, 0x0a63195a, L"eof"}, |
- {TOKfor, 0x0a7d67a7, L"for"}, |
- {TOKnan, 0x0b4f91dd, L"nan"}, |
- {TOKksnot, 0x0b4fd9b1, L"not"}, |
- {TOKvar, 0x0c2203e9, L"var"}, |
- {TOKthen, 0x2d5738cf, L"then"}, |
- {TOKelse, 0x45f65ee9, L"else"}, |
- {TOKexit, 0x4731d6ba, L"exit"}, |
- {TOKdownto, 0x4caadc3b, L"downto"}, |
- {TOKreturn, 0x4db8bd60, L"return"}, |
- {TOKinfinity, 0x5c0a010a, L"infinity"}, |
- {TOKendwhile, 0x5c64bff0, L"endwhile"}, |
- {TOKforeach, 0x67e31f38, L"foreach"}, |
- {TOKendfunc, 0x68f984a3, L"endfunc"}, |
- {TOKelseif, 0x78253218, L"elseif"}, |
- {TOKwhile, 0x84229259, L"while"}, |
- {TOKendfor, 0x8ab49d7e, L"endfor"}, |
- {TOKthrow, 0x8db05c94, L"throw"}, |
- {TOKstep, 0xa7a7887c, L"step"}, |
- {TOKupto, 0xb5155328, L"upto"}, |
- {TOKcontinue, 0xc0340685, L"continue"}, |
- {TOKfunc, 0xcdce60ec, L"func"}, |
- {TOKendif, 0xe0e8fee6, L"endif"}, |
-}; |
- |
-const XFA_FM_TOKEN KEYWORD_START = TOKdo; |
-const XFA_FM_TOKEN KEYWORD_END = TOKendif; |
- |
-} // namespace |
- |
-const FX_WCHAR* XFA_FM_KeywordToString(XFA_FM_TOKEN op) { |
- if (op < KEYWORD_START || op > KEYWORD_END) |
- return L""; |
- return keyWords[op].m_keyword; |
-} |
- |
-CXFA_FMToken::CXFA_FMToken() : m_type(TOKreserver), m_uLinenum(1) {} |
- |
-CXFA_FMToken::CXFA_FMToken(FX_DWORD uLineNum) |
- : m_type(TOKreserver), m_uLinenum(uLineNum) {} |
- |
-CXFA_FMLexer::CXFA_FMLexer(const CFX_WideStringC& wsFormCalc, |
- CXFA_FMErrorInfo* pErrorInfo) |
- : m_ptr(wsFormCalc.GetPtr()), m_uCurrentLine(1), m_pErrorInfo(pErrorInfo) {} |
- |
-CXFA_FMToken* CXFA_FMLexer::NextToken() { |
- m_pToken.reset(Scan()); |
- return m_pToken.get(); |
-} |
- |
-CXFA_FMToken* CXFA_FMLexer::Scan() { |
- uint16_t ch = 0; |
- CXFA_FMToken* p = new CXFA_FMToken(m_uCurrentLine); |
- if (!XFA_FMDChar::isAvalid(m_ptr)) { |
- ch = XFA_FMDChar::get(m_ptr); |
- Error(FMERR_UNSUPPORTED_CHAR, ch); |
- return p; |
- } |
- int iRet = 0; |
- while (1) { |
- if (!XFA_FMDChar::isAvalid(m_ptr)) { |
- ch = XFA_FMDChar::get(m_ptr); |
- Error(FMERR_UNSUPPORTED_CHAR, ch); |
- return p; |
- } |
- ch = XFA_FMDChar::get(m_ptr); |
- switch (ch) { |
- case 0: |
- p->m_type = TOKeof; |
- return p; |
- case 0x0A: |
- ++m_uCurrentLine; |
- p->m_uLinenum = m_uCurrentLine; |
- XFA_FMDChar::inc(m_ptr); |
- break; |
- case 0x0D: |
- XFA_FMDChar::inc(m_ptr); |
- break; |
- case ';': { |
- const FX_WCHAR* pTemp = 0; |
- Comment(m_ptr, pTemp); |
- m_ptr = pTemp; |
- } break; |
- case '"': { |
- const FX_WCHAR* pTemp = 0; |
- p->m_type = TOKstring; |
- iRet = String(p, m_ptr, pTemp); |
- m_ptr = pTemp; |
- } |
- return p; |
- case '0': |
- case '1': |
- case '2': |
- case '3': |
- case '4': |
- case '5': |
- case '6': |
- case '7': |
- case '8': |
- case '9': { |
- p->m_type = TOKnumber; |
- const FX_WCHAR* pTemp = 0; |
- iRet = Number(p, m_ptr, pTemp); |
- m_ptr = pTemp; |
- if (iRet) { |
- Error(FMERR_BAD_SUFFIX_NUMBER); |
- return p; |
- } |
- } |
- return p; |
- case '=': |
- XFA_FMDChar::inc(m_ptr); |
- if (XFA_FMDChar::isAvalid(m_ptr)) { |
- ch = XFA_FMDChar::get(m_ptr); |
- if (ch == '=') { |
- p->m_type = TOKeq; |
- XFA_FMDChar::inc(m_ptr); |
- return p; |
- } else { |
- p->m_type = TOKassign; |
- return p; |
- } |
- } else { |
- ch = XFA_FMDChar::get(m_ptr); |
- Error(FMERR_UNSUPPORTED_CHAR, ch); |
- return p; |
- } |
- break; |
- case '<': |
- XFA_FMDChar::inc(m_ptr); |
- if (XFA_FMDChar::isAvalid(m_ptr)) { |
- ch = XFA_FMDChar::get(m_ptr); |
- if (ch == '=') { |
- p->m_type = TOKle; |
- XFA_FMDChar::inc(m_ptr); |
- return p; |
- } else if (ch == '>') { |
- p->m_type = TOKne; |
- XFA_FMDChar::inc(m_ptr); |
- return p; |
- } else { |
- p->m_type = TOKlt; |
- return p; |
- } |
- } else { |
- ch = XFA_FMDChar::get(m_ptr); |
- Error(FMERR_UNSUPPORTED_CHAR, ch); |
- return p; |
- } |
- break; |
- case '>': |
- XFA_FMDChar::inc(m_ptr); |
- if (XFA_FMDChar::isAvalid(m_ptr)) { |
- ch = XFA_FMDChar::get(m_ptr); |
- if (ch == '=') { |
- p->m_type = TOKge; |
- XFA_FMDChar::inc(m_ptr); |
- return p; |
- } else { |
- p->m_type = TOKgt; |
- return p; |
- } |
- } else { |
- ch = XFA_FMDChar::get(m_ptr); |
- Error(FMERR_UNSUPPORTED_CHAR, ch); |
- return p; |
- } |
- break; |
- case ',': |
- p->m_type = TOKcomma; |
- XFA_FMDChar::inc(m_ptr); |
- return p; |
- case '(': |
- p->m_type = TOKlparen; |
- XFA_FMDChar::inc(m_ptr); |
- return p; |
- case ')': |
- p->m_type = TOKrparen; |
- XFA_FMDChar::inc(m_ptr); |
- return p; |
- case '[': |
- p->m_type = TOKlbracket; |
- XFA_FMDChar::inc(m_ptr); |
- return p; |
- case ']': |
- p->m_type = TOKrbracket; |
- XFA_FMDChar::inc(m_ptr); |
- return p; |
- case '&': |
- XFA_FMDChar::inc(m_ptr); |
- p->m_type = TOKand; |
- return p; |
- case '|': |
- XFA_FMDChar::inc(m_ptr); |
- p->m_type = TOKor; |
- return p; |
- case '+': |
- XFA_FMDChar::inc(m_ptr); |
- p->m_type = TOKplus; |
- return p; |
- case '-': |
- XFA_FMDChar::inc(m_ptr); |
- p->m_type = TOKminus; |
- return p; |
- case '*': |
- XFA_FMDChar::inc(m_ptr); |
- p->m_type = TOKmul; |
- return p; |
- case '/': |
- XFA_FMDChar::inc(m_ptr); |
- if (XFA_FMDChar::isAvalid(m_ptr)) { |
- ch = XFA_FMDChar::get(m_ptr); |
- if (ch == '/') { |
- const FX_WCHAR* pTemp = 0; |
- Comment(m_ptr, pTemp); |
- m_ptr = pTemp; |
- break; |
- } else { |
- p->m_type = TOKdiv; |
- return p; |
- } |
- } else { |
- ch = XFA_FMDChar::get(m_ptr); |
- Error(FMERR_UNSUPPORTED_CHAR, ch); |
- return p; |
- } |
- break; |
- case '.': |
- XFA_FMDChar::inc(m_ptr); |
- if (XFA_FMDChar::isAvalid(m_ptr)) { |
- ch = XFA_FMDChar::get(m_ptr); |
- if (ch == '.') { |
- p->m_type = TOKdotdot; |
- XFA_FMDChar::inc(m_ptr); |
- return p; |
- } else if (ch == '*') { |
- p->m_type = TOKdotstar; |
- XFA_FMDChar::inc(m_ptr); |
- return p; |
- } else if (ch == '#') { |
- p->m_type = TOKdotscream; |
- XFA_FMDChar::inc(m_ptr); |
- return p; |
- } else if (ch <= '9' && ch >= '0') { |
- p->m_type = TOKnumber; |
- const FX_WCHAR* pTemp = 0; |
- XFA_FMDChar::dec(m_ptr); |
- iRet = Number(p, m_ptr, pTemp); |
- m_ptr = pTemp; |
- if (iRet) { |
- Error(FMERR_BAD_SUFFIX_NUMBER); |
- } |
- return p; |
- } else { |
- p->m_type = TOKdot; |
- return p; |
- } |
- } else { |
- ch = XFA_FMDChar::get(m_ptr); |
- Error(FMERR_UNSUPPORTED_CHAR, ch); |
- return p; |
- } |
- case 0x09: |
- case 0x0B: |
- case 0x0C: |
- case 0x20: |
- XFA_FMDChar::inc(m_ptr); |
- break; |
- default: { |
- const FX_WCHAR* pTemp = 0; |
- iRet = Identifiers(p, m_ptr, pTemp); |
- m_ptr = pTemp; |
- if (iRet) { |
- return p; |
- } |
- p->m_type = IsKeyword(p->m_wstring); |
- } |
- return p; |
- } |
- } |
-} |
- |
-FX_DWORD CXFA_FMLexer::Number(CXFA_FMToken* t, |
- const FX_WCHAR* p, |
- const FX_WCHAR*& pEnd) { |
- FX_DOUBLE number = 0; |
- if (XFA_FMDChar::string2number(p, &number, pEnd)) { |
- return 1; |
- } |
- if (pEnd && XFA_FMDChar::isAlpha(pEnd)) { |
- return 1; |
- } |
- t->m_wstring = CFX_WideStringC(p, (pEnd - p)); |
- return 0; |
-} |
- |
-FX_DWORD CXFA_FMLexer::String(CXFA_FMToken* t, |
- const FX_WCHAR* p, |
- const FX_WCHAR*& pEnd) { |
- const FX_WCHAR* pStart = p; |
- uint16_t ch = 0; |
- XFA_FMDChar::inc(p); |
- ch = XFA_FMDChar::get(p); |
- while (ch) { |
- if (!XFA_FMDChar::isAvalid(p)) { |
- ch = XFA_FMDChar::get(p); |
- pEnd = p; |
- t->m_wstring = CFX_WideStringC(pStart, (pEnd - pStart)); |
- Error(FMERR_UNSUPPORTED_CHAR, ch); |
- return 1; |
- } |
- if (ch == '"') { |
- XFA_FMDChar::inc(p); |
- if (!XFA_FMDChar::isAvalid(p)) { |
- ch = XFA_FMDChar::get(p); |
- pEnd = p; |
- t->m_wstring = CFX_WideStringC(pStart, (pEnd - pStart)); |
- Error(FMERR_UNSUPPORTED_CHAR, ch); |
- return 1; |
- } |
- ch = XFA_FMDChar::get(p); |
- if (ch == '"') { |
- goto NEXT; |
- } else { |
- break; |
- } |
- } |
- NEXT: |
- XFA_FMDChar::inc(p); |
- ch = XFA_FMDChar::get(p); |
- } |
- pEnd = p; |
- t->m_wstring = CFX_WideStringC(pStart, (pEnd - pStart)); |
- return 0; |
-} |
- |
-FX_DWORD CXFA_FMLexer::Identifiers(CXFA_FMToken* t, |
- const FX_WCHAR* p, |
- const FX_WCHAR*& pEnd) { |
- const FX_WCHAR* pStart = p; |
- uint16_t ch = 0; |
- ch = XFA_FMDChar::get(p); |
- XFA_FMDChar::inc(p); |
- if (!XFA_FMDChar::isAvalid(p)) { |
- pEnd = p; |
- t->m_wstring = CFX_WideStringC(pStart, (pEnd - pStart)); |
- Error(FMERR_UNSUPPORTED_CHAR, ch); |
- return 1; |
- } |
- ch = XFA_FMDChar::get(p); |
- while (ch) { |
- if (!XFA_FMDChar::isAvalid(p)) { |
- pEnd = p; |
- t->m_wstring = CFX_WideStringC(pStart, (pEnd - pStart)); |
- Error(FMERR_UNSUPPORTED_CHAR, ch); |
- return 1; |
- } |
- ch = XFA_FMDChar::get(p); |
- if (XFA_FMDChar::isUnicodeAlpha(ch)) { |
- XFA_FMDChar::inc(p); |
- } else { |
- pEnd = p; |
- t->m_wstring = CFX_WideStringC(pStart, (pEnd - pStart)); |
- return 0; |
- } |
- } |
- pEnd = p; |
- t->m_wstring = CFX_WideStringC(pStart, (pEnd - pStart)); |
- return 0; |
-} |
- |
-void CXFA_FMLexer::Comment(const FX_WCHAR* p, const FX_WCHAR*& pEnd) { |
- unsigned ch = 0; |
- XFA_FMDChar::inc(p); |
- ch = XFA_FMDChar::get(p); |
- while (ch) { |
- if (ch == 0x0D) { |
- XFA_FMDChar::inc(p); |
- pEnd = p; |
- return; |
- } |
- if (ch == 0x0A) { |
- ++m_uCurrentLine; |
- XFA_FMDChar::inc(p); |
- pEnd = p; |
- return; |
- } |
- XFA_FMDChar::inc(p); |
- ch = XFA_FMDChar::get(p); |
- } |
- pEnd = p; |
-} |
- |
-XFA_FM_TOKEN CXFA_FMLexer::IsKeyword(const CFX_WideStringC& str) { |
- int32_t iLength = str.GetLength(); |
- uint32_t uHash = FX_HashCode_String_GetW(str.GetPtr(), iLength, TRUE); |
- int32_t iStart = KEYWORD_START, iEnd = KEYWORD_END; |
- int32_t iMid = (iStart + iEnd) / 2; |
- XFA_FMKeyword keyword; |
- do { |
- iMid = (iStart + iEnd) / 2; |
- keyword = keyWords[iMid]; |
- if (uHash == keyword.m_uHash) { |
- return keyword.m_type; |
- } else if (uHash < keyword.m_uHash) { |
- iEnd = iMid - 1; |
- } else { |
- iStart = iMid + 1; |
- } |
- } while (iStart <= iEnd); |
- return TOKidentifier; |
-} |
- |
-void CXFA_FMLexer::Error(XFA_FM_ERRMSG msg, ...) { |
- m_pErrorInfo->linenum = m_uCurrentLine; |
- const FX_WCHAR* lpMessageInfo = XFA_FM_ErrorMsg(msg); |
- va_list ap; |
- va_start(ap, msg); |
- m_pErrorInfo->message.FormatV(lpMessageInfo, ap); |
- va_end(ap); |
-} |
- |
-FX_BOOL CXFA_FMLexer::HasError() const { |
- if (m_pErrorInfo->message.IsEmpty()) { |
- return FALSE; |
- } |
- return TRUE; |
-} |