Here is a small experimentation I did. It is known that parsing of
very large files is slowed down by the fact that we do a whole lot
of string compares instead of a good old binary search like in lyxlex.
Typically, when loading a big file from Herbert,
Buffer::parseSingleLyXformat2Token takes 25.6% of the time,
and a big part of that is due to 2114004 strings comparisons.
It occured to me that this could be alleviated by reordering the tokens
in the chained if:s. I whipped the following simple script
--counttokens----
#!/bin/sh
grep '^\\' $1 | sed -e 's/^\(\\[^ ]*\) .*$/\1/'|sort|uniq -c|sort -k1,1nr
--------
And running it on the book, I get
ginette: ./counttokens fichiers/Versuch.lyx
7767 \layout
6239 \begin_inset
6239 \end_inset
5138 \latex
4841 \backslash
2638 \emph
2555 \family
2540 \noun
2537 \color
2534 \bar
2534 \shape
2420 \size
1733 \series
277 \begin_float
277 \end_float
[..]
Therefore I moved the fonts, insets and backslash (useful for ERT) tests
to the top of the method.
The result is that Buffer::parseSingleLyXformat2Token takes 16.2% of the
time, and the 404503 string comparisons now take less time than
InsertChar, which makes more sense.
I attach the patch instead of commiting it, because I want to have
a virtual nod from Lars before. I would not want to break the carefully
crafted compatibility reading...
Note that this is tested with lyxstring. Results with normal strings
may be completely different...
JMarc
Index: src/ChangeLog
===================================================================
RCS file: /usr/local/lyx/cvsroot/lyx-devel/src/ChangeLog,v
retrieving revision 1.651
diff -u -r1.651 ChangeLog
--- src/ChangeLog 27 Mar 2002 23:27:12 -0000 1.651
+++ src/ChangeLog 27 Mar 2002 23:29:00 -0000
@@ -1,3 +1,9 @@
+2002-03-28 Jean-Marc Lasgouttes <[EMAIL PROTECTED]>
+
+ * buffer.C (parseSingleLyXformat2Token): reorder a bit the tests
+ in order to reduce drastically the number of comparisons needed to
+ parse a large document
+
2002-03-27 Jean-Marc Lasgouttes <[EMAIL PROTECTED]>
* lyxfunc.C (getStatus): return 'disabled' early for LFUN_NOACTION
Index: src/buffer.C
===================================================================
RCS file: /usr/local/lyx/cvsroot/lyx-devel/src/buffer.C,v
retrieving revision 1.322
diff -u -r1.322 buffer.C
--- src/buffer.C 25 Mar 2002 11:15:25 -0000 1.322
+++ src/buffer.C 27 Mar 2002 23:29:11 -0000
@@ -564,6 +564,102 @@
#endif
#ifndef NO_COMPABILITY
+ } else if (token == "\\end_inset") {
+ lyxerr << "Solitary \\end_inset. Missing \\begin_inset?.\n"
+ << "Last inset read was: " << last_inset_read
+ << endl;
+ // Simply ignore this. The insets do not have
+ // to read this.
+ // But insets should read it, it is a part of
+ // the inset isn't it? Lgb.
+ } else if (token == "\\begin_inset") {
+#ifndef NO_COMPABILITY
+ insertErtContents(par, pos, false);
+ ert_stack.push(ert_comp);
+ ert_comp = ErtComp();
+#endif
+ readInset(lex, par, pos, font);
+#ifndef NO_COMPABILITY
+ ert_comp = ert_stack.top();
+ ert_stack.pop();
+ insertErtContents(par, pos);
+#endif
+ } else if (token == "\\family") {
+ lex.next();
+ font.setLyXFamily(lex.getString());
+ } else if (token == "\\series") {
+ lex.next();
+ font.setLyXSeries(lex.getString());
+ } else if (token == "\\shape") {
+ lex.next();
+ font.setLyXShape(lex.getString());
+ } else if (token == "\\size") {
+ lex.next();
+ font.setLyXSize(lex.getString());
+#ifndef NO_COMPABILITY
+ } else if (token == "\\latex") {
+ lex.next();
+ string const tok = lex.getString();
+ if (tok == "no_latex") {
+ // Do the insetert.
+ insertErtContents(par, pos);
+ } else if (tok == "latex") {
+ ert_comp.active = true;
+ ert_comp.font = font;
+ } else if (tok == "default") {
+ // Do the insetert.
+ insertErtContents(par, pos);
+ } else {
+ lex.printError("Unknown LaTeX font flag "
+ "`$$Token'");
+ }
+#endif
+ } else if (token == "\\lang") {
+ lex.next();
+ string const tok = lex.getString();
+ Language const * lang = languages.getLanguage(tok);
+ if (lang) {
+ font.setLanguage(lang);
+ } else {
+ font.setLanguage(params.language);
+ lex.printError("Unknown language `$$Token'");
+ }
+ } else if (token == "\\numeric") {
+ lex.next();
+ font.setNumber(font.setLyXMisc(lex.getString()));
+ } else if (token == "\\emph") {
+ lex.next();
+ font.setEmph(font.setLyXMisc(lex.getString()));
+ } else if (token == "\\bar") {
+ lex.next();
+ string const tok = lex.getString();
+ // This is dirty, but gone with LyX3. (Asger)
+ if (tok == "under")
+ font.setUnderbar(LyXFont::ON);
+ else if (tok == "no")
+ font.setUnderbar(LyXFont::OFF);
+ else if (tok == "default")
+ font.setUnderbar(LyXFont::INHERIT);
+ else
+ lex.printError("Unknown bar font flag "
+ "`$$Token'");
+ } else if (token == "\\noun") {
+ lex.next();
+ font.setNoun(font.setLyXMisc(lex.getString()));
+ } else if (token == "\\color") {
+ lex.next();
+ font.setLyXColor(lex.getString());
+ } else if (token == "\\backslash") {
+#ifndef NO_COMPABILITY
+ if (ert_comp.active) {
+ ert_comp.contents += "\\";
+ } else {
+#endif
+ par->insertChar(pos, '\\', font);
+ ++pos;
+#ifndef NO_COMPABILITY
+ }
+#endif
} else if (token == "\\begin_float") {
insertErtContents(par, pos);
//insertErtContents(par, pos, false);
@@ -951,71 +1047,6 @@
} else if (token == "\\float_placement") {
lex.nextToken();
params.float_placement = lex.getString();
- } else if (token == "\\family") {
- lex.next();
- font.setLyXFamily(lex.getString());
- } else if (token == "\\series") {
- lex.next();
- font.setLyXSeries(lex.getString());
- } else if (token == "\\shape") {
- lex.next();
- font.setLyXShape(lex.getString());
- } else if (token == "\\size") {
- lex.next();
- font.setLyXSize(lex.getString());
-#ifndef NO_COMPABILITY
- } else if (token == "\\latex") {
- lex.next();
- string const tok = lex.getString();
- if (tok == "no_latex") {
- // Do the insetert.
- insertErtContents(par, pos);
- } else if (tok == "latex") {
- ert_comp.active = true;
- ert_comp.font = font;
- } else if (tok == "default") {
- // Do the insetert.
- insertErtContents(par, pos);
- } else {
- lex.printError("Unknown LaTeX font flag "
- "`$$Token'");
- }
-#endif
- } else if (token == "\\lang") {
- lex.next();
- string const tok = lex.getString();
- Language const * lang = languages.getLanguage(tok);
- if (lang) {
- font.setLanguage(lang);
- } else {
- font.setLanguage(params.language);
- lex.printError("Unknown language `$$Token'");
- }
- } else if (token == "\\numeric") {
- lex.next();
- font.setNumber(font.setLyXMisc(lex.getString()));
- } else if (token == "\\emph") {
- lex.next();
- font.setEmph(font.setLyXMisc(lex.getString()));
- } else if (token == "\\bar") {
- lex.next();
- string const tok = lex.getString();
- // This is dirty, but gone with LyX3. (Asger)
- if (tok == "under")
- font.setUnderbar(LyXFont::ON);
- else if (tok == "no")
- font.setUnderbar(LyXFont::OFF);
- else if (tok == "default")
- font.setUnderbar(LyXFont::INHERIT);
- else
- lex.printError("Unknown bar font flag "
- "`$$Token'");
- } else if (token == "\\noun") {
- lex.next();
- font.setNoun(font.setLyXMisc(lex.getString()));
- } else if (token == "\\color") {
- lex.next();
- font.setLyXColor(lex.getString());
} else if (token == "\\align") {
int tmpret = lex.findToken(string_align);
if (tmpret == -1) ++tmpret;
@@ -1062,26 +1093,6 @@
lex.eatLine();
par->params().labelWidthString(lex.getString());
// do not delete this token, it is still needed!
- } else if (token == "\\end_inset") {
- lyxerr << "Solitary \\end_inset. Missing \\begin_inset?.\n"
- << "Last inset read was: " << last_inset_read
- << endl;
- // Simply ignore this. The insets do not have
- // to read this.
- // But insets should read it, it is a part of
- // the inset isn't it? Lgb.
- } else if (token == "\\begin_inset") {
-#ifndef NO_COMPABILITY
- insertErtContents(par, pos, false);
- ert_stack.push(ert_comp);
- ert_comp = ErtComp();
-#endif
- readInset(lex, par, pos, font);
-#ifndef NO_COMPABILITY
- ert_comp = ert_stack.top();
- ert_stack.pop();
- insertErtContents(par, pos);
-#endif
} else if (token == "\\SpecialChar") {
LyXLayout const & layout =
textclasslist[params.textclass][par->layout()];
@@ -1157,17 +1168,6 @@
par->bibkey = new InsetBibKey(p);
}
par->bibkey->read(this, lex);
- } else if (token == "\\backslash") {
-#ifndef NO_COMPABILITY
- if (ert_comp.active) {
- ert_comp.contents += "\\";
- } else {
-#endif
- par->insertChar(pos, '\\', font);
- ++pos;
-#ifndef NO_COMPABILITY
- }
-#endif
} else if (token == "\\the_end") {
#ifndef NO_COMPABILITY
// If we still have some ert active here we have to insert
