fpc/ide/whtml.pas
2009-08-15 21:02:22 +00:00

918 lines
30 KiB
ObjectPascal
Raw Blame History

This file contains invisible Unicode characters

This file contains invisible Unicode characters that are indistinguishable to humans but may be processed differently by a computer. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

{
This file is part of the Free Pascal Integrated Development Environment
Copyright (c) 1998 by Berczi Gabor
See the file COPYING.FPC, included in this distribution,
for details about the copyright.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
**********************************************************************}
unit WHTML;
{$I globdir.inc}
interface
uses Objects;
type
PTextFile = ^TTextFile;
TTextFile = object(TObject)
function GetLine(Idx: sw_integer; var S: string): boolean; virtual;
function GetFileName : string; virtual;
end;
PMemoryTextFile = ^TMemoryTextFile;
TMemoryTextFile = object(TTextFile)
constructor Init;
procedure AddLine(const S: string); virtual;
function GetLine(Idx: sw_integer; var S: string): boolean; virtual;
function GetFileName : string; virtual;
function GetLineCount : sw_integer;
destructor Done; virtual;
private
Lines : PUnsortedStrCollection;
end;
PDOSTextFile = ^TDOSTextFile;
TDOSTextFile = object(TMemoryTextFile)
constructor Init(AFileName: string);
function GetFileName : string; virtual;
private
DosFileName : string;
end;
PSGMLParser = ^TSGMLParser;
TSGMLParser = object(TObject)
constructor Init;
function Process(HTMLFile: PTextFile): boolean; virtual;
function ProcessLine(LineText: string): boolean; virtual;
destructor Done; virtual;
public
Line,LinePos: sw_integer;
procedure DocSoftBreak; virtual;
function GetFileName : string;
function DocAddTextChar(C: char): boolean; virtual;
procedure DocAddText(S: string); virtual;
procedure DocProcessTag(Tag: string); virtual;
procedure DocProcessComment(Comment: string); virtual;
function DocDecodeNamedEntity(Name: string; var Entity: string): boolean; virtual;
private
CurTag: string;
FileName : string;
InTag,InComment,InString: boolean;
end;
PHTMLParser = ^THTMLParser;
THTMLParser = object(TSGMLParser)
procedure DocSoftBreak; virtual;
function DocAddTextChar(C: char): boolean; virtual;
procedure DocProcessTag(Tag: string); virtual;
function DocGetTagParam(Name: string; var Value: string): boolean; virtual;
procedure DocProcessComment(Comment: string); virtual;
function DocDecodeNamedEntity(Name: string; var E: string): boolean; virtual;
public
TagName,TagParams: string;
DisableCrossIndexing : boolean;
procedure DocUnknownTag; virtual;
procedure DocTYPE; virtual;
procedure DocHTML(Entered: boolean); virtual;
procedure DocHEAD(Entered: boolean); virtual;
procedure DocMETA; virtual;
procedure DocTITLE(Entered: boolean); virtual;
procedure DocBODY(Entered: boolean); virtual;
procedure DocAnchor(Entered: boolean); virtual;
procedure DocHeading(Level: integer; Entered: boolean); virtual;
procedure DocParagraph(Entered: boolean); virtual;
procedure DocBreak; virtual;
procedure DocImage; virtual;
procedure DocBold(Entered: boolean); virtual;
procedure DocCite(Entered: boolean); virtual;
procedure DocCode(Entered: boolean); virtual;
procedure DocEmphasized(Entered: boolean); virtual;
procedure DocItalic(Entered: boolean); virtual;
procedure DocKbd(Entered: boolean); virtual;
procedure DocPreformatted(Entered: boolean); virtual;
procedure DocSample(Entered: boolean); virtual;
procedure DocStrong(Entered: boolean); virtual;
procedure DocTeleType(Entered: boolean); virtual;
procedure DocVariable(Entered: boolean); virtual;
procedure DocSpan(Entered: boolean); virtual;
procedure DocDiv(Entered: boolean); virtual;
procedure DocList(Entered: boolean); virtual;
procedure DocOrderedList(Entered: boolean); virtual;
procedure DocListItem(Entered: boolean); virtual;
procedure DocDefList(Entered: boolean); virtual;
procedure DocDefTerm(Entered: boolean); virtual;
procedure DocDefExp(Entered: boolean); virtual;
procedure DocTable(Entered: boolean); virtual;
procedure DocTableRow(Entered: boolean); virtual;
procedure DocTableHeaderItem(Entered: boolean); virtual;
procedure DocTableItem(Entered: boolean); virtual;
procedure DocHorizontalRuler; virtual;
end;
Type
PTopicLinkCollection = ^TTopicLinkCollection;
TTopicLinkCollection = object(TStringCollection)
procedure Insert(Item: Pointer); virtual;
function At(Index: sw_Integer): PString;
function AddItem(Item: string): sw_integer;
end;
function EncodeHTMLCtx(FileID: integer; LinkNo: word): longint;
procedure DecodeHTMLCtx(Ctx: longint; var FileID: word; var LinkNo: word);
implementation
uses
WUtils;
function TTextFile.GetLine(Idx: sw_integer; var S: string): boolean;
begin
Abstract;
GetLine:=false;
end;
function TTextFile.GetFileName : string;
begin
GetFileName:='unknown';
end;
constructor TMemoryTextFile.Init;
begin
inherited Init;
New(Lines, Init(500,500));
end;
function TMemoryTextFile.GetFileName : string;
begin
GetFileName:='unknown';
end;
function TMemoryTextFile.GetLineCount : sw_integer;
begin
GetLineCount:=Lines^.Count;
end;
procedure TMemoryTextFile.AddLine(const S: string);
begin
Lines^.Insert(NewStr(S));
end;
function TMemoryTextFile.GetLine(Idx: sw_integer; var S: string): boolean;
var OK: boolean;
PS: PString;
begin
OK:=(Lines<>nil) and (Idx<Lines^.Count);
if OK then
begin
PS:=Lines^.At(Idx);
if PS=nil then S:='' else S:=PS^;
end;
GetLine:=OK;
end;
destructor TMemoryTextFile.Done;
begin
if Lines<>nil then
Dispose(Lines, Done);
Lines:=nil;
inherited Done;
end;
constructor TDOSTextFile.Init(AFileName: string);
var f: file;
linecomplete,hasCR: boolean;
S: string;
OldFMode : Integer;
begin
inherited Init;
if AFileName='' then Fail;
{$I-}
Assign(f,AFileName);
OldFMode:= FileMode;
FileMode:= 0;
Reset(f,1);
FileMode:= OldFMode;
if IOResult<>0 then Fail;
DosFileName:=AFileName;
Dispose(Lines,Done);
New(Lines, Init(500,2000));
while (Eof(f)=false) and (IOResult=0) do
begin
ReadlnFromFile(f,S,linecomplete,hasCR,true);
AddLine(S);
end;
Close(f);
{$I+}
end;
function TDosTextFile.GetFileName : string;
begin
GetFileName:=DosFileName;
end;
constructor TSGMLParser.Init;
begin
inherited Init;
FileName:='';
end;
function TSGMLParser.GetFileName : string;
begin
GetFileName:=FileName;
end;
function TSGMLParser.Process(HTMLFile: PTextFile): boolean;
var S: string;
OK,LineOK: boolean;
begin
if HTMLFile=nil then Exit;
InTag:=false; InComment:=false; InString:=false; CurTag:='';
Line:=0; OK:=true;
FileName:=HTMLFile^.GetFileName;
repeat
LineOK:=HTMLFile^.GetLine(Line,S);
if LineOK then
begin
Inc(Line);
OK:=ProcessLine(S);
end;
until (LineOK=false) or (OK=false);
Process:=OK;
end;
function TSGMLParser.ProcessLine(LineText: string): boolean;
var OK: boolean;
C: char;
NewInString: boolean;
OldInComment: boolean;
WasThereAnyText: boolean;
Pos2: integer;
Name,Entity: string;
LiteralCode: boolean;
LiteralStart,LiteralEnd,P: integer;
const TabSize : integer = 8;
Tab = #9;
begin
WasThereAnyText:=false;
OK:=true; LinePos:=1;
LiteralStart:=0; LiteralEnd:=0;
repeat
P:=Pos(TAB,LineText);
if P>0 then
LineText:=copy(LineText,1,P-1)+CharStr(' ',TabSize)+copy(LineText,P+1,255);
until P=0;
while (LinePos<=length(LineText)) and OK do
begin
LiteralCode:=false;
NewInString:=InString; OldInComment:=InComment;
C:=LineText[LinePos];
LiteralCode:=(LiteralStart<=LinePos) and (LinePos<=LiteralEnd);
if (LiteralCode=false) and (C='&') then
begin
LiteralStart:=0; LiteralEnd:=0;
Name:=''; Pos2:=LinePos+1;
while (Pos2<=length(LineText)) and (LineText[Pos2]<>';') do
begin
Name:=Name+LineText[Pos2];
Inc(Pos2);
end;
Inc(Pos2);
if DocDecodeNamedEntity(Name,Entity) then
begin
LineText:=copy(LineText,1,LinePos-1)+Entity+copy(LineText,Pos2,255);
LiteralStart:=LinePos; LiteralEnd:=LiteralStart+length(Entity)-1;
C:=LineText[LinePos];
end;
end;
LiteralCode:=(LiteralStart<=LinePos) and (LinePos<=LiteralEnd);
if (LiteralCode=false) and (C='"') and (InTag=true) and (InString=false) then
NewInString:=true;
if (LiteralCode=false) and (C='<') and (InTag=false) then
InTag:=true;
if InTag then CurTag:=CurTag+C else
WasThereAnyText:=DocAddTextChar(C);
if (LiteralCode=false) and InTag and (InString=false) and (CurTag='<!--') then
InComment:=true;
{ A comment can be longer than 255 chars
move the test to LineText string,
This is why the Previous, Next and Up Tags where not working ... PM
if (LiteralCode=false) and InTag and InComment and (InString=false) and (length(CurTag)>=3) and
(copy(CurTag,length(CurTag)-2,3)='-->') then
InComment:=false; }
if (LiteralCode=false) and InTag and InComment and (InString=false) and (LinePos>=3) and
(copy(LineText,LinePos-2,3)='-->') then
InComment:=false;
if (LiteralCode=false) and (C='"') and (InTag=true) and (InString=true) then
NewInString:=false;
if (LiteralCode=false) and (C='>') and (InTag=true) then
begin
InTag:=false;
if OldInComment then
DocProcessComment(CurTag)
else
DocProcessTag(CurTag);
CurTag:='';
end;
InString:=NewInString;
Inc(LinePos);
end;
{ whtml does not depend on whelp,
so I can not use hscLineBreak here. PM }
if InTag then
begin
if InString then
CurTag:=CurTag+#0;
end
else if WasThereAnyText then
DocSoftBreak;
ProcessLine:=true;
end;
procedure TSGMLParser.DocSoftBreak;
begin
Abstract;
end;
function TSGMLParser.DocAddTextChar(C: char): boolean;
begin
Abstract;
DocAddTextChar:=false;
end;
procedure TSGMLParser.DocAddText(S: string);
var I: sw_integer;
begin
for I:=1 to length(S) do
DocAddTextChar(S[I]);
end;
function TSGMLParser.DocDecodeNamedEntity(Name: string; var Entity: string): boolean;
begin
DocDecodeNamedEntity:=false;
end;
procedure TSGMLParser.DocProcessTag(Tag: string);
begin
Abstract;
end;
procedure TSGMLParser.DocProcessComment(Comment: string);
begin
Abstract;
end;
destructor TSGMLParser.Done;
begin
inherited Done;
end;
procedure THTMLParser.DocSoftBreak;
begin
end;
function THTMLParser.DocAddTextChar(C: char): boolean;
begin
{ Abstract }
DocAddTextChar:=false;
end;
function THTMLParser.DocDecodeNamedEntity(Name: string; var E: string): boolean;
var Found: boolean;
Code: word;
CC: word;
begin
Found:=true; Code:=$ffff;
Name:=LowCaseStr(Name);
if copy(Name,1,1)='#' then
begin
if Name[2]='x' then
Val('$'+copy(Name,3,255),Code,CC)
else
Val(copy(Name,2,255),Code,CC);
if CC<>0 then
begin
{$ifdef DEBUG}
DebugMessage(FileName,'NamedEntity '+Name+' not converted',1,1);
{$endif DEBUG}
Code:=$ffff;
end;
end;
{ #0 to #127 is same for Unicode and Code page 437 }
if (code<=127) then
begin
E:=chr(code);
DocDecodeNamedEntity:=true;
exit;
end;
if (Code=$22{34}) or (Name='quot') then E:='"' else { double quote sign }
if (Code=$26{38}) or (Name='amp') then E:='&' else { ampersand }
if (Code=$27{39}) or (Name='apos') then E:='''' else { apostrophe }
if (Code=$3C{60}) or (Name='lt') then E:='<' else { less-than sign }
if (Code=$3E{62}) or (Name='gt') then E:='>' else { greater-than sign }
if (Code=$5B) then E:='[' else { [ }
if (Code=$5C) then E:='\' else { \ }
if (Code=$5D) then E:=']' else { ] }
if (Code=$5E) then E:='^' else { ^ }
if (Code=$5F) then E:='_' else { _ }
if (Code=160) or (Name='nbsp') then E:=#255 else { no-break space }
if (Code=161) or (Name='iexcl') then E:='<27>' else { inverted exclamation mark }
if (Code=162) or (Name='cent') then E:='<27>' else { cent sign }
if (Code=163) or (Name='pound') then E:='<27>' else { pound sterling sign }
if (Code=164) or (Name='curren') then E:='$' else { general currency sign }
if (Code=165) or (Name='yen') then E:='' else { yen sign }
if (Code=166) or (Name='brvbar') then E:='|' else { broken vertical bar }
if (Code=167) or (Name='sect') then E:='' else { section sign }
if (Code=168) or (Name='uml') then E:='"' else { umlaut (dieresis) }
if (Code=169) or (Name='copy') then E:='(C)' else { copyright sign }
(* if (Code=170) or (Name='ordf') then E:=#255 else { ordinal indicator, feminine }*)
if (Code=171) or (Name='laquo') then E:='"' else { angle quotation mark -left }
if (Code=172) or (Name='not') then E:='!' else { not sign }
if (Code=173) or (Name='shy') then E:='-' else { soft hypen }
if (Code=174) or (Name='reg') then E:='(R)' else { registered sign }
(* if (Code=175) or (Name='macr') then E:='?' else { macron }*)
if (Code=176) or (Name='deg') then E:='<27>' else { degree sign }
if (Code=177) or (Name='plusmn') then E:='<27>' else { plus-or-minus sign }
if (Code=178) or (Name='sup2') then E:='<27>' else { superscript 2 }
if (Code=179) or (Name='sup3') then E:='^3' else { superscript 3 }
if (Code=180) or (Name='acute') then E:='''' else { acute accent }
if (Code=181) or (Name='micro') then E:='<27>' else { micro sign }
(* if (Code=182) or (Name='para') then E:='?' else { paragraph sign }*)
if (Code=183) or (Name='middot') then E:='<27>' else { middle dot }
(* if (Code=184) or (Name='cedil') then E:='?' else { cedilla }*)
if (Code=185) or (Name='sup1') then E:='^1' else { superscript 1 }
(* if (Code=186) or (Name='ordm') then E:='?' else { ordinal indicator, masculine }*)
if (Code=187) or (Name='raquo') then E:='"' else { angle quoatation mark -right }
if (Code=188) or (Name='frac14') then E:='<27>' else { fraction one-quarter }
if (Code=189) or (Name='frac12') then E:='<27>' else { fraction one-half }
if (Code=190) or (Name='frac34') then E:='3/4' else { fraction three-quarters }
if (Code=191) or (Name='iquest') then E:='<27>' else { inverted question mark }
if (Code=192) or (Name='Agrave') then E:='A' else { capital A, grave accent }
if (Code=193) or (Name='Aacute') then E:='A' else { capital A, acute accent }
if (Code=194) or (Name='Acirc') then E:='A' else { capital A, circumflex accent }
if (Code=195) or (Name='Atilde') then E:='A' else { capital A, tilde accent }
if (Code=196) or (Name='Auml') then E:='<27>' else { capital A, dieresis or umlaut }
if (Code=197) or (Name='Aring') then E:='' else { capital A, ring }
if (Code=198) or (Name='AElig') then E:='<27>' else { capital AE diphthong }
if (Code=199) or (Name='Ccedil') then E:='<27>' else { capital C, cedilla }
if (Code=200) or (Name='Egrave') then E:='' else { capital E, grave accent }
if (Code=201) or (Name='Eacute') then E:='' else { capital E, acute accent }
if (Code=202) or (Name='Ecirc') then E:='E' else { capital E, circumflex accent }
if (Code=203) or (Name='Euml') then E:='E' else { capital E, dieresis or umlaut }
if (Code=204) or (Name='Igrave') then E:='I' else { capital I, grave accent }
if (Code=205) or (Name='Iacute') then E:='I' else { capital I, acute accent }
if (Code=206) or (Name='Icirc') then E:='I' else { capital I, circumflex accent }
if (Code=207) or (Name='Iuml') then E:='I' else { capital I, dieresis or umlaut }
(* if (Code=208) or (Name='ETH') then E:='?' else { capital Eth, Icelandic }*)
if (Code=209) or (Name='Ntidle') then E:='<27>' else { capital N, tilde }
if (Code=210) or (Name='Ograve') then E:='O' else { capital O, grave accent }
if (Code=211) or (Name='Oacute') then E:='O' else { capital O, acute accent }
if (Code=212) or (Name='Ocirc') then E:='O' else { capital O, circumflex accent }
if (Code=213) or (Name='Otilde') then E:='O' else { capital O, tilde }
if (Code=214) or (Name='Ouml') then E:='<27>' else { capital O, dieresis or umlaut }
if (Code=215) or (Name='times') then E:='*' else { multiply sign }
if (Code=216) or (Name='Oslash') then E:='O' else { capital O, slash }
if (Code=217) or (Name='Ugrave') then E:='U' else { capital U, grave accent }
if (Code=218) or (Name='Uacute') then E:='U' else { capital U, acute accent }
if (Code=219) or (Name='Ucirc') then E:='U' else { capital U, circumflex accent }
if (Code=220) or (Name='Uuml') then E:='<27>' else { capital U, dieresis or umlaut }
if (Code=221) or (Name='Yacute') then E:='Y' else { capital Y, acute accent }
(* if (Code=222) or (Name='THORN') then E:='?' else { capital THORN, Icelandic }*)
if (Code=223) or (Name='szlig') then E:='<27>' else { small sharp S, German }
if (Code=224) or (Name='agrave') then E:='<27>' else { small a, grave accent }
if (Code=225) or (Name='aacute') then E:='<27>' else { small a, acute accent }
if (Code=226) or (Name='acirc') then E:='<27>' else { small a, circumflex accent }
if (Code=227) or (Name='atilde') then E:='<27>' else { small a, tilde }
if (Code=228) or (Name='auml') then E:='<27>' else { small a, dieresis or umlaut }
if (Code=229) or (Name='aring') then E:='<27>' else { small a, ring }
if (Code=230) or (Name='aelig') then E:='ae' else { small ae, diphthong }
if (Code=231) or (Name='ccedil') then E:='<27>' else { small c, cedilla }
if (Code=232) or (Name='egrave') then E:='<27>' else { small e, grave accent }
if (Code=233) or (Name='eacute') then E:='<27>' else { small e, acute accent }
if (Code=234) or (Name='ecirc') then E:='<27>' else { small e, circumflex accent }
if (Code=235) or (Name='euml') then E:='<27>' else { small e, dieresis or umlaut }
if (Code=236) or (Name='igrave') then E:='' else { small i, grave accent }
if (Code=237) or (Name='iacute') then E:='<27>' else { small i, acute accent }
if (Code=238) or (Name='icirc') then E:='<27>' else { small i, circumflex accent }
if (Code=239) or (Name='iuml') then E:='<27>' else { small i, dieresis or umlaut }
(* if (Code=240) or (Name='eth') then E:='?' else { small eth, Icelandic }*)
if (Code=241) or (Name='ntilde') then E:='<27>' else { small n, tilde }
if (Code=242) or (Name='ograve') then E:='<27>' else { small o, grave accent }
if (Code=243) or (Name='oacute') then E:='<27>' else { small o, acute accent }
if (Code=244) or (Name='ocirc') then E:='<27>' else { small o, circumflex accent }
if (Code=245) or (Name='otilde') then E:='<27>' else { small o, tilde }
if (Code=246) or (Name='ouml') then E:='<27>' else { small o, dieresis or umlaut }
if (Code=247) or (Name='divide') then E:='/' else { divide sign }
if (Code=248) or (Name='oslash') then E:='"' else { small o, slash }
if (Code=249) or (Name='ugrave') then E:='<27>' else { small u, grave accent }
if (Code=250) or (Name='uacute') then E:='<27>' else { small u, acute accent }
if (Code=251) or (Name='ucirc') then E:='<27>' else { small u, circumflex accent }
if (Code=252) or (Name='uuml') then E:='' else { small u, dieresis or umlaut }
if (Code=253) or (Name='yacute') then E:='y' else { small y, acute accent }
(* if (Code=254) or (Name='thorn') then E:='?' else { small thorn, Icelandic }*)
if (Code=255) or (Name='yuml') then E:='y' else { small y, dieresis or umlaut }
{ Special codes appearing in TeXH generated files }
if (code=$2c6{710}) or (Name='circ') then E:='^' else { Modifier Letter Circumflex Accent }
if (code=$2dc{732}) or (Name='tilde') then E:='~' else { Small tilde }
if (code=$2013{8211}) or (Name='endash') then E:='-' else { En dash }
if (code=$2014{8212}) or (Name='emdash') then E:='--' else { Em dash }
if (Code=$2018{8216}) or (Name='lsquo') then E:='`' else { Acute accent as generated by TeXH }
if (Code=$2019{8217}) or (Name='rsquo') then E:='''' else { acute accent as generated by TeXH }
if (code=$201C{8220}) or (Name='ldquo') then E:='''''' else { left double quotation marks }
if (code=$201D{8221}) or (Name='rdquo') then E:='``' else { right double quotation marks }
if (code=$2026{8230}) or (Name='hellip') then E:='...' else { horizontal ellipsis }
if (Code=$FB00) then E:='ff' else { ff together }
if (Code=$FB01) then E:='fi' else { fi together }
if (Code=$FB02) then E:='fl' else { fl together }
if (Code=$FB03) then E:='ffi' else { ffi together }
if (Code=$FB04) then E:='ffl' else { ffl together }
Found:=false;
DocDecodeNamedEntity:=Found;
{$ifdef DEBUG}
if (Code<>$ffff) and not found then
begin
DebugMessage(FileName,'NamedEntity '+Name+' not handled',1,1);
end;
{$endif DEBUG}
end;
procedure THTMLParser.DocProcessTag(Tag: string);
var UTagName,ETagName: string[30];
P: byte;
NotEndTag: boolean;
begin
if copy(Tag,1,1)='<' then Delete(Tag,1,1);
if copy(Tag,length(Tag),1)='>' then Delete(Tag,length(Tag),1);
Tag:=Trim(Tag);
P:=Pos(' ',Tag); if P=0 then P:=length(Tag)+1;
TagName:=copy(Tag,1,P-1); TagParams:=copy(Tag,P+1,255);
UTagName:=UpcaseStr(TagName);
NotEndTag:=copy(TagName,1,1)<>'/';
if NotEndTag then ETagName:=UTagName else ETagName:=copy(UTagName,2,255);
{ <BR/> is also a Break tag... }
if Copy(ETagName,Length(ETagName),1)='/' then
begin
ETagName:=copy(ETagName,1,Length(ETagName)-1);
NotEndTag:=false;
end;
if (UTagName='!DOCTYPE') then DocTYPE else
{ Section tags }
if (ETagName='HTML') then DocHTML(NotEndTag) else
if (ETagName='HEAD') then DocHEAD(NotEndTag) else
if (ETagName='TITLE') then DocTITLE(NotEndTag) else
if (ETagName='BODY') then DocBODY(NotEndTag) else
{ Anchor tags }
if (ETagName='A') then DocAnchor(NotEndTag) else
{ Direct formatting directives }
if (ETagName='H1') then DocHeading(1,NotEndTag) else
if (ETagName='H2') then DocHeading(2,NotEndTag) else
if (ETagName='H3') then DocHeading(3,NotEndTag) else
if (ETagName='H4') then DocHeading(4,NotEndTag) else
if (ETagName='H5') then DocHeading(5,NotEndTag) else
if (ETagName='H6') then DocHeading(6,NotEndTag) else
if (ETagName='P') then DocParagraph(NotEndTag) else
if (ETagName='BR') then DocBreak else
if (ETagName='B') then DocBold(NotEndTag) else
if (ETagName='CITE') then DocCite(NotEndTag) else
if (ETagName='CODE') then DocCode(NotEndTag) else
if (ETagName='EM') then DocEmphasized(NotEndTag) else
if (ETagName='I') then DocItalic(NotEndTag) else
if (ETagName='KBD') then DocKbd(NotEndTag) else
if (ETagName='PRE') then DocPreformatted(NotEndTag) else
if (ETagName='SAMP') then DocSample(NotEndTag) else
if (ETagName='STRONG') then DocStrong(NotEndTag) else
if (ETagName='TT') then DocTeleType(NotEndTag) else
if (ETagName='VAR') then DocVariable(NotEndTag) else
if (ETagName='SPAN') then DocSpan(NotEndTag) else
if (ETagName='DIV') then DocDiv(NotEndTag) else
{ Unordered & ordered lists }
if (ETagName='UL') then DocList(NotEndTag) else
if (ETagName='OL') then DocOrderedList(NotEndTag) else
if (ETagName='LI') then DocListItem(NotEndTag) else
{ Definition list }
if (ETagName='DL') then DocDefList(NotEndTag) else
if (ETagName='DT') then DocDefTerm(NotEndTag) else
if (ETagName='DD') then DocDefExp(NotEndTag) else
{ Table }
if (ETagName='TABLE') then DocTable(NotEndTag) else
if (ETagName='TR') then DocTableRow(NotEndTag) else
if (ETagName='TH') then DocTableHeaderItem(NotEndTag) else
if (ETagName='TD') then DocTableItem(NotEndTag) else
{ Misc. tags }
if (UTagName='META') then DocMETA else
if (UTagName='IMG') then DocImage else
if (UTagName='HR') then DocHorizontalRuler else
DocUnknownTag;
end;
function THTMLParser.DocGetTagParam(Name: string; var Value: string): boolean;
var Found: boolean;
S: string;
ParamName,ParamValue: string;
InStr: boolean;
I: sw_integer;
begin
Found:=false;
Name:=UpcaseStr(Name);
Value:='';
S:=TagParams;
repeat
InStr:=false;
ParamName:=''; ParamValue:='';
S:=Trim(S); I:=1;
while (I<=length(S)) and (S[I]<>'=') do
begin
if S[I]=' ' then
ParamName:=''
else
ParamName:=ParamName+S[I];
Inc(I);
end;
ParamName:=Trim(ParamName);
if S[I]='=' then
begin
Inc(I); InStr:=false;
while (I<=length(S)) and (S[I]=' ') do
Inc(I);
if (I<=length(S)) and (S[I]='"') then
begin
InStr:=true;
Inc(I);
end;
while (I<=length(S)) and ((InStr=true) or (S[I]<>' ')) do
begin
if S[I]='"' then
begin
InStr:=not InStr;
if InStr=false then Break;
end
else
ParamValue:=ParamValue+S[I];
Inc(I);
end;
end;
Found:=(Name=UpcaseStr(ParamName));
if Found then Value:=ParamValue;
Delete(S,1,I);
until Found or (S='');
DocGetTagParam:=Found;
end;
procedure THTMLParser.DocProcessComment(Comment: string);
begin
end;
procedure THTMLParser.DocUnknownTag;
begin
end;
procedure THTMLParser.DocTYPE;
begin
end;
procedure THTMLParser.DocHTML(Entered: boolean);
begin
end;
procedure THTMLParser.DocHEAD(Entered: boolean);
begin
end;
procedure THTMLParser.DocMETA;
begin
end;
procedure THTMLParser.DocTITLE(Entered: boolean);
begin
end;
procedure THTMLParser.DocBODY(Entered: boolean);
begin
end;
procedure THTMLParser.DocAnchor(Entered: boolean);
begin
end;
procedure THTMLParser.DocHeading(Level: integer; Entered: boolean);
begin
end;
procedure THTMLParser.DocParagraph(Entered: boolean);
begin
end;
procedure THTMLParser.DocBreak;
begin
end;
procedure THTMLParser.DocImage;
begin
end;
procedure THTMLParser.DocBold(Entered: boolean);
begin
end;
procedure THTMLParser.DocCite(Entered: boolean);
begin
end;
procedure THTMLParser.DocCode(Entered: boolean);
begin
end;
procedure THTMLParser.DocEmphasized(Entered: boolean);
begin
end;
procedure THTMLParser.DocItalic(Entered: boolean);
begin
end;
procedure THTMLParser.DocKbd(Entered: boolean);
begin
end;
procedure THTMLParser.DocPreformatted(Entered: boolean);
begin
end;
procedure THTMLParser.DocSample(Entered: boolean);
begin
end;
procedure THTMLParser.DocStrong(Entered: boolean);
begin
end;
procedure THTMLParser.DocTeleType(Entered: boolean);
begin
end;
procedure THTMLParser.DocVariable(Entered: boolean);
begin
end;
procedure THTMLParser.DocSpan(Entered: boolean);
begin
end;
procedure THTMLParser.DocDiv(Entered: boolean);
var
S: String;
begin
if Entered then
begin
if DocGetTagParam('CLASS',S) then
if S='crosslinks' then
begin
DisableCrossIndexing:=true;
{$ifdef DEBUG}
DebugMessage(GetFileName,'Crosslinks found',Line,LinePos);
{$endif DEBUG}
end;
end
else
begin
{$ifdef DEBUG}
if DisableCrossIndexing then
begin
DebugMessage(GetFileName,'Crosslinks end found',Line,LinePos);
end;
{$endif DEBUG}
DisableCrossIndexing:=false;
end;
end;
procedure THTMLParser.DocList(Entered: boolean);
begin
end;
procedure THTMLParser.DocOrderedList(Entered: boolean);
begin
end;
procedure THTMLParser.DocListItem(Entered: boolean);
begin
end;
procedure THTMLParser.DocDefList(Entered: boolean);
begin
end;
procedure THTMLParser.DocDefTerm(Entered: boolean);
begin
end;
procedure THTMLParser.DocDefExp(Entered: boolean);
begin
end;
procedure THTMLParser.DocTable(Entered: boolean);
var
S: String;
begin
if Entered then
begin
if DocGetTagParam('CLASS',S) then
if S='bar' then
begin
DisableCrossIndexing:=true;
{$ifdef DEBUG}
DebugMessage(GetFileName,'Bar table found, cross indexing disabled ',Line,LinePos);
{$endif DEBUG}
end;
end
else
begin
{$ifdef DEBUG}
if DisableCrossIndexing then
begin
DebugMessage(GetFileName,'Bar table end found',Line,LinePos);
end;
{$endif DEBUG}
DisableCrossIndexing:=false;
end;
end;
procedure THTMLParser.DocTableRow(Entered: boolean);
begin
end;
procedure THTMLParser.DocTableHeaderItem(Entered: boolean);
begin
end;
procedure THTMLParser.DocTableItem(Entered: boolean);
begin
end;
procedure THTMLParser.DocHorizontalRuler;
begin
end;
function EncodeHTMLCtx(FileID: integer; LinkNo: word): longint;
var Ctx: longint;
begin
Ctx:=(longint(FileID) shl 16)+LinkNo;
EncodeHTMLCtx:=Ctx;
end;
procedure DecodeHTMLCtx(Ctx: longint; var FileID: word; var LinkNo: word);
begin
if (Ctx shr 16)=0 then
begin
FileID:=$ffff; LinkNo:=0;
end
else
begin
FileID:=Ctx shr 16; LinkNo:=Ctx and $ffff;
end;
end;
procedure TTopicLinkCollection.Insert(Item: Pointer);
begin
AtInsert(Count,Item);
end;
function TTopicLinkCollection.At(Index: sw_Integer): PString;
begin
At:=inherited At(Index);
end;
function TTopicLinkCollection.AddItem(Item: string): sw_integer;
var Idx: sw_integer;
begin
if Item='' then Idx:=-1 else
if Search(@Item,Idx)=false then
begin
AtInsert(Count,NewStr(Item));
Idx:=Count-1;
end;
AddItem:=Idx;
end;
END.