John P. Linderman wrote:
Using what is to me the more obvious [0-9] pattern takes almost 50 times as
long as using the [[:digit:]] pattern. Seems very strange.
Thanks for reporting that. In general, patterns like [a-z] can be much slower
than [[:lower:]] due to poorly-thought-out POSIX interfaces. However, [0-9] is a
special case: we can optimize such patterns safely if both ends are ASCII
digits. I installed the attached patch to Gnulib to do that; it fixes the
performance glitch you noticed, at least for me.
>From 6afba02d7869d39ed7f61981045ddbdcb2814101 Mon Sep 17 00:00:00 2001
From: Paul Eggert <egg...@cs.ucla.edu>
Date: Tue, 21 Mar 2017 19:05:17 -0700
Subject: [PATCH] dfa: make [0-9] faster in non-C locales
Problem reported by John P. Linderman (Bug#26193).
* lib/dfa.c (parse_bracket_exp): Remove redundant assignment.
If both ends of the range are ASCII digits, do not worry about
multi-character collating sequences and the like. Be consistent
about using isalpha as a precondition for setbit_case_fold_c.
---
ChangeLog | 9 +++++++++
lib/dfa.c | 35 ++++++++++++-----------------------
2 files changed, 21 insertions(+), 23 deletions(-)
diff --git a/ChangeLog b/ChangeLog
index 625f007..33bd6a1 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,3 +1,12 @@
+2017-03-21 Paul Eggert <egg...@cs.ucla.edu>
+
+ dfa: make [0-9] faster in non-C locales
+ Problem reported by John P. Linderman (Bug#26193).
+ * lib/dfa.c (parse_bracket_exp): Remove redundant assignment.
+ If both ends of the range are ASCII digits, do not worry about
+ multi-character collating sequences and the like. Be consistent
+ about using isalpha as a precondition for setbit_case_fold_c.
+
2017-03-19 Bruno Haible <br...@clisp.org>
lock: Fix compilation error with HP-UX IA64 cc.
diff --git a/lib/dfa.c b/lib/dfa.c
index 5bac288..e97dae1c 100644
--- a/lib/dfa.c
+++ b/lib/dfa.c
@@ -551,8 +551,9 @@ struct dfa
bool, size_t *, bool *);
/* The locale is simple, like the C locale. These locales can be
- processed more efficiently, e.g., the relationship between lower-
- and upper-case letters is 1-1. */
+ processed more efficiently, as they are single-byte, their native
+ character set is in collating-sequence order, and they do not
+ have multi-character collating elements. */
bool simple_locale;
/* Other cached information derived from the locale. */
@@ -1012,7 +1013,6 @@ parse_bracket_exp (struct dfa *dfa)
if (invert)
{
c = bracket_fetch_wc (dfa);
- invert = true;
known_bracket_exp = dfa->simple_locale;
}
wint_t wc = dfa->lex.wctok;
@@ -1143,24 +1143,14 @@ parse_bracket_exp (struct dfa *dfa)
/* Treat [x-y] as a range if x != y. */
if (wc != wc2 || wc == WEOF)
{
- if (dfa->localeinfo.multibyte)
- known_bracket_exp = false;
- else if (dfa->simple_locale)
+ if (dfa->simple_locale
+ || (isasciidigit (c) & isasciidigit (c2)))
{
- int ci;
- for (ci = c; ci <= c2; ci++)
- setbit (ci, &ccl);
- if (dfa->syntax.case_fold)
- {
- int uc = toupper (c);
- int uc2 = toupper (c2);
- for (ci = 0; ci < NOTCHAR; ci++)
- {
- int uci = toupper (ci);
- if (uc <= uci && uci <= uc2)
- setbit (ci, &ccl);
- }
- }
+ for (int ci = c; ci <= c2; ci++)
+ if (dfa->syntax.case_fold && isalpha (ci))
+ setbit_case_fold_c (ci, &ccl);
+ else
+ setbit (ci, &ccl);
}
else
known_bracket_exp = false;
@@ -1174,7 +1164,7 @@ parse_bracket_exp (struct dfa *dfa)
if (!dfa->localeinfo.multibyte)
{
- if (dfa->syntax.case_fold)
+ if (dfa->syntax.case_fold && isalpha (c))
setbit_case_fold_c (c, &ccl);
else
setbit (c, &ccl);
@@ -1209,7 +1199,7 @@ parse_bracket_exp (struct dfa *dfa)
if (! known_bracket_exp)
return BACKREF;
- if (dfa->localeinfo.multibyte)
+ if (dfa->localeinfo.multibyte && (invert || dfa->lex.brack.nchars != 0))
{
dfa->lex.brack.invert = invert;
dfa->lex.brack.cset = emptyset (&ccl) ? -1 : charclass_index (dfa, &ccl);
@@ -1218,7 +1208,6 @@ parse_bracket_exp (struct dfa *dfa)
if (invert)
{
- assert (!dfa->localeinfo.multibyte);
notset (&ccl);
if (dfa->syntax.syntax_bits & RE_HAT_LISTS_NOT_NEWLINE)
clrbit ('\n', &ccl);
--
2.7.4
