Speed up git-grep by using PCRE v2 as a backend
cover Speed up git-grep by using PCRE v2 as a backend
Thought I'd send this to the list too. This is first of my WIP post-PCRE v2 inclusion series's.
In addition to the huge speed improvements for grep -P noted in the culmination of that series[1], this speeds up all other types of grep invocations (fixed string & POSIX basic/extended) by using an experimental PCRE API to translate those patterns to PCRE syntax.
Fixed string grep is sped up by ~15-50%, and any greps containing regexes by 40-70%, with 50% seeming to be the average for most normal patterns.
It isn't ready for the reasons noted in the last patch in the series, and currently brings most of PCRE into git in compat/pcre2 since it uses an experimental API.
The 5/6 patch is pretty much ready though and works on stock PCRE, it fixes TODO tests for patterns that contain a \0, and enables regex metacharacters in such patterns (right now they're all implicitly fixed strings), see the discussion in that patch for some of the caveats.
That patch will most likely be dropped by the list, it can be retrieved from https://github.com/avar/git
avar/grep-and-pcre-and-more, or the whole series viewed at
https://github.com/git/git/compare/master...avar:avar/grep-and-pcre-and-more.
1. <[email protected]>
(https://public-inbox.org/git/[email protected]/)
Ævar Arnfjörð Bjarmason (6):
Makefile & compat/pcre2: add ability to build an embedded PCRE Makefile & compat/pcre2: add dependency on pcre2_convert.c compat/pcre2: import pcre2 from svn trunk
test-lib: add LIBPCRE1 & LIBPCRE2 prerequisites
grep: support regex patterns containing \0 via PCRE v2
grep: use PCRE v2 under the hood for -G & -E for amazing speedup
Makefile | 53 + compat/pcre2/get-pcre2.sh | 68 + compat/pcre2/src/pcre2.h | 832 ++ compat/pcre2/src/pcre2_auto_possess.c | 1291 ++ compat/pcre2/src/pcre2_chartables.c | 1 + compat/pcre2/src/pcre2_chartables.c.dist | 198 + compat/pcre2/src/pcre2_compile.c | 9626 +++++++++++++++ compat/pcre2/src/pcre2_config.c | 222 + compat/pcre2/src/pcre2_context.c | 450 + compat/pcre2/src/pcre2_convert.c | 724 ++ compat/pcre2/src/pcre2_error.c | 327 + compat/pcre2/src/pcre2_find_bracket.c | 218 + compat/pcre2/src/pcre2_internal.h | 1967 +++ compat/pcre2/src/pcre2_intmodedep.h | 884 ++ compat/pcre2/src/pcre2_jit_compile.c | 12307 +++++++++++++++++++ compat/pcre2/src/pcre2_jit_match.c | 189 + compat/pcre2/src/pcre2_jit_misc.c | 227 + compat/pcre2/src/pcre2_maketables.c | 157 + compat/pcre2/src/pcre2_match.c | 6826 ++++++++++ compat/pcre2/src/pcre2_match_data.c | 147 + compat/pcre2/src/pcre2_newline.c | 243 + compat/pcre2/src/pcre2_ord2utf.c | 120 + compat/pcre2/src/pcre2_string_utils.c | 201 + compat/pcre2/src/pcre2_study.c | 1624 +++ compat/pcre2/src/pcre2_tables.c | 765 ++ compat/pcre2/src/pcre2_ucd.c | 3761 ++++++ compat/pcre2/src/pcre2_ucp.h | 268 + compat/pcre2/src/pcre2_valid_utf.c | 398 + compat/pcre2/src/pcre2_xclass.c | 271 + compat/pcre2/src/sljit/sljitConfig.h | 145 + compat/pcre2/src/sljit/sljitConfigInternal.h | 725 ++ compat/pcre2/src/sljit/sljitExecAllocator.c | 312 + compat/pcre2/src/sljit/sljitLir.c | 2224 ++++ compat/pcre2/src/sljit/sljitLir.h | 1392 +++ compat/pcre2/src/sljit/sljitNativeARM_32.c | 2326 ++++ compat/pcre2/src/sljit/sljitNativeARM_64.c | 2104 ++++ compat/pcre2/src/sljit/sljitNativeARM_T2_32.c | 1987 +++ compat/pcre2/src/sljit/sljitNativeMIPS_32.c | 437 + compat/pcre2/src/sljit/sljitNativeMIPS_64.c | 539 + compat/pcre2/src/sljit/sljitNativeMIPS_common.c | 2110 ++++ compat/pcre2/src/sljit/sljitNativePPC_32.c | 276 + compat/pcre2/src/sljit/sljitNativePPC_64.c | 447 + compat/pcre2/src/sljit/sljitNativePPC_common.c | 2421 ++++ compat/pcre2/src/sljit/sljitNativeSPARC_32.c | 165 + compat/pcre2/src/sljit/sljitNativeSPARC_common.c | 1471 +++ compat/pcre2/src/sljit/sljitNativeTILEGX-encoder.c | 10159 +++++++++++++++ compat/pcre2/src/sljit/sljitNativeTILEGX_64.c | 2555 ++++ compat/pcre2/src/sljit/sljitNativeX86_32.c | 602 + compat/pcre2/src/sljit/sljitNativeX86_64.c | 742 ++ compat/pcre2/src/sljit/sljitNativeX86_common.c | 2921 +++++ compat/pcre2/src/sljit/sljitProtExecAllocator.c | 421 + compat/pcre2/src/sljit/sljitUtils.c | 334 + grep.c | 73 +- grep.h | 5 + t/README | 18 + t/t7008-grep-binary.sh | 87 +- t/test-lib.sh | 3 + 57 files changed, 81335 insertions(+), 31 deletions(-) create mode 100755 compat/pcre2/get-pcre2.sh create mode 100644 compat/pcre2/src/pcre2.h create mode 100644 compat/pcre2/src/pcre2_auto_possess.c create mode 120000 compat/pcre2/src/pcre2_chartables.c create mode 100644 compat/pcre2/src/pcre2_chartables.c.dist create mode 100644 compat/pcre2/src/pcre2_compile.c create mode 100644 compat/pcre2/src/pcre2_config.c create mode 100644 compat/pcre2/src/pcre2_context.c create mode 100644 compat/pcre2/src/pcre2_convert.c create mode 100644 compat/pcre2/src/pcre2_error.c create mode 100644 compat/pcre2/src/pcre2_find_bracket.c create mode 100644 compat/pcre2/src/pcre2_internal.h create mode 100644 compat/pcre2/src/pcre2_intmodedep.h create mode 100644 compat/pcre2/src/pcre2_jit_compile.c create mode 100644 compat/pcre2/src/pcre2_jit_match.c create mode 100644 compat/pcre2/src/pcre2_jit_misc.c create mode 100644 compat/pcre2/src/pcre2_maketables.c create mode 100644 compat/pcre2/src/pcre2_match.c create mode 100644 compat/pcre2/src/pcre2_match_data.c create mode 100644 compat/pcre2/src/pcre2_newline.c create mode 100644 compat/pcre2/src/pcre2_ord2utf.c create mode 100644 compat/pcre2/src/pcre2_string_utils.c create mode 100644 compat/pcre2/src/pcre2_study.c create mode 100644 compat/pcre2/src/pcre2_tables.c create mode 100644 compat/pcre2/src/pcre2_ucd.c create mode 100644 compat/pcre2/src/pcre2_ucp.h create mode 100644 compat/pcre2/src/pcre2_valid_utf.c create mode 100644 compat/pcre2/src/pcre2_xclass.c create mode 100644 compat/pcre2/src/sljit/sljitConfig.h create mode 100644 compat/pcre2/src/sljit/sljitConfigInternal.h create mode 100644 compat/pcre2/src/sljit/sljitExecAllocator.c create mode 100644 compat/pcre2/src/sljit/sljitLir.c create mode 100644 compat/pcre2/src/sljit/sljitLir.h create mode 100644 compat/pcre2/src/sljit/sljitNativeARM_32.c create mode 100644 compat/pcre2/src/sljit/sljitNativeARM_64.c create mode 100644 compat/pcre2/src/sljit/sljitNativeARM_T2_32.c create mode 100644 compat/pcre2/src/sljit/sljitNativeMIPS_32.c create mode 100644 compat/pcre2/src/sljit/sljitNativeMIPS_64.c create mode 100644 compat/pcre2/src/sljit/sljitNativeMIPS_common.c create mode 100644 compat/pcre2/src/sljit/sljitNativePPC_32.c create mode 100644 compat/pcre2/src/sljit/sljitNativePPC_64.c create mode 100644 compat/pcre2/src/sljit/sljitNativePPC_common.c create mode 100644 compat/pcre2/src/sljit/sljitNativeSPARC_32.c create mode 100644 compat/pcre2/src/sljit/sljitNativeSPARC_common.c create mode 100644 compat/pcre2/src/sljit/sljitNativeTILEGX-encoder.c create mode 100644 compat/pcre2/src/sljit/sljitNativeTILEGX_64.c create mode 100644 compat/pcre2/src/sljit/sljitNativeX86_32.c create mode 100644 compat/pcre2/src/sljit/sljitNativeX86_64.c create mode 100644 compat/pcre2/src/sljit/sljitNativeX86_common.c create mode 100644 compat/pcre2/src/sljit/sljitProtExecAllocator.c create mode 100644 compat/pcre2/src/sljit/sljitUtils.c
Signature · 1 line
2.11.0
PATCH 1/6 Makefile & compat/pcre2: add ability to build an embedded PCRE
Add a USE_LIBPCRE2_BUNDLED=YesIHaveNoPackagedVersion flag to the Makefile which'll use the PCRE v2 shipped in compat/pcre2 instead of trying to find it via -lpcre2-8 on the installed system.
As covered in a previous commits ("grep: add support for PCRE v2", 2017-04-08) there are major benefits to using a bleeding edge PCRE v2, but more importantly I'd like to experiment with making PCRE a mandatory dependency to power various internal features of grep/log without the user being aware that they're using the library under the hood, similar to how we use kwset now for fixed-string searches.
Imposing that hard dependency on everyone using git would bother a lot of people, whereas if git itself ships PCRE it's no more bothersome than the code using kwset, i.e. it can be invisible to the builder & user, and allow git to target newer PCRE APIs without worrying about versioning.
See [1] for a mostly one-sided pcre-dev mailing list thread discussing how embed the library.
Implementation details:
* I configured PCRE v2 with ./configure --enable-jit --enable-utf
* It sets a lot of -DHAVE_* but these are used by the subset of the files I copied, many are either unused or only used by pcre2test.c which isn't brought in by the script.
* These -DHAVE_* flags are something we have already by default & assume in other git.git code, so it should be fine to define it.
* -DPCRE2_CODE_UNIT_WIDTH=8 only compiles the functions linking to -lpcre2-8 would have gotten us.
* All the limits / sizes are the PCRE defaults, the MATCH_LIMIT_RECURSION define is a synonym for MATCH_LIMIT_DEPTH in older versions, it allows building against older (currently release) versions of the library.
* -DNEWLINE_DEFAULT=2 means only \n is recognized as a newline. This corresponds to the --enable-newline-is-lf option. It's also possible to set this to CR, CRLF, any of CR, LF, or CRLF, or any Unicode newline character being recognized as \n.
This *might* have to be customized on Windows, but I think the grep machinery always splits on newlines for us already, so this probably works on Windows as-is, but needs testing.
1. https://lists.exim.org/lurker/thread/20170507.223619.fbee8f00.en.html
--- Makefile | 52 ++++++++++++++++++++++++++++++++++++ compat/pcre2/get-pcre2.sh | 67 +++++++++++++++++++++++++++++++++++++++++++++++ 2 files changed, 119 insertions(+) create mode 100755 compat/pcre2/get-pcre2.sh
Makefile +52 -0
@@ -34,6 +34,11 @@ all:: # library. The USE_LIBPCRE flag will likely be changed to mean v2 by # default in future releases. # +# Define USE_LIBPCRE2_BUNDLED=YesIHaveNoPackagedVersion in addition to +# USE_LIBPCRE2=YesPlease if you'd like to use a copy of PCRE version 2 +# bunded with Git. This is for setups where getting a hold of a +# packaged PCRE is inconvenient. +# # Define LIBPCREDIR=/foo/bar if your PCRE header and library files are in # /foo/bar/include and /foo/bar/lib directories. #
@@ -1105,8 +1110,10 @@ endif ifdef USE_LIBPCRE2 BASIC_CFLAGS += -DUSE_LIBPCRE2 +ifndef USE_LIBPCRE2_BUNDLED EXTLIBS += -lpcre2-8 endif +endif ifdef LIBPCREDIR BASIC_CFLAGS += -I$(LIBPCREDIR)/include
@@ -1505,6 +1512,50 @@ ifdef NO_REGEX COMPAT_CFLAGS += -Icompat/regex COMPAT_OBJS += compat/regex/regex.o endif +ifdef USE_LIBPCRE2_BUNDLED +ifndef USE_LIBPCRE2 +$(error please set USE_LIBPCRE2=YesPlease when setting \ +USE_LIBPCRE2_BUNDLED=$(USE_LIBPCRE2_BUNDLED)) +endif + COMPAT_CFLAGS += \ + -Icompat/pcre2/src \ + -DHAVE_BCOPY=1 \ + -DHAVE_INTTYPES_H=1 \ + -DHAVE_MEMMOVE=1 \ + -DHAVE_STDINT_H=1 \ + -DPCRE2_CODE_UNIT_WIDTH=8 \ + -DLINK_SIZE=2 \ + -DHEAP_LIMIT=20000000 \ + -DMATCH_LIMIT=10000000 \ + -DMATCH_LIMIT_DEPTH=10000000 \ + -DMATCH_LIMIT_RECURSION=10000000 \ + -DMAX_NAME_COUNT=10000 \ + -DMAX_NAME_SIZE=32 \ + -DPARENS_NEST_LIMIT=250 \ + -DNEWLINE_DEFAULT=2 \ + -DSUPPORT_JIT \ + -DSUPPORT_UNICODE + COMPAT_OBJS += \ + compat/pcre2/src/pcre2_auto_possess.o \ + compat/pcre2/src/pcre2_chartables.o \ + compat/pcre2/src/pcre2_compile.o \ + compat/pcre2/src/pcre2_config.o \ + compat/pcre2/src/pcre2_context.o \ + compat/pcre2/src/pcre2_error.o \ + compat/pcre2/src/pcre2_find_bracket.o \ + compat/pcre2/src/pcre2_jit_compile.o \ + compat/pcre2/src/pcre2_maketables.o \ + compat/pcre2/src/pcre2_match.o \ + compat/pcre2/src/pcre2_match_data.o \ + compat/pcre2/src/pcre2_newline.o \ + compat/pcre2/src/pcre2_ord2utf.o \ + compat/pcre2/src/pcre2_string_utils.o \ + compat/pcre2/src/pcre2_study.o \ + compat/pcre2/src/pcre2_tables.o \ + compat/pcre2/src/pcre2_ucd.o \ + compat/pcre2/src/pcre2_valid_utf.o \ + compat/pcre2/src/pcre2_xclass.o +endif ifdef NATIVE_CRLF BASIC_CFLAGS += -DNATIVE_CRLF endif
@@ -2259,6 +2310,7 @@ GIT-BUILD-OPTIONS: FORCE @echo NO_EXPAT=\''$(subst ','\'',$(subst ','\'',$(NO_EXPAT)))'\' >>$@+ @echo USE_LIBPCRE1=\''$(subst ','\'',$(subst ','\'',$(USE_LIBPCRE)))'\' >>$@+ @echo USE_LIBPCRE2=\''$(subst ','\'',$(subst ','\'',$(USE_LIBPCRE2)))'\' >>$@+ + @echo USE_LIBPCRE2_BUNDLED=\''$(subst ','\'',$(subst ','\'',$(USE_LIBPCRE2_BUNDLED)))'\' >>$@+ @echo NO_PERL=\''$(subst ','\'',$(subst ','\'',$(NO_PERL)))'\' >>$@+ @echo NO_PTHREADS=\''$(subst ','\'',$(subst ','\'',$(NO_PTHREADS)))'\' >>$@+ @echo NO_PYTHON=\''$(subst ','\'',$(subst ','\'',$(NO_PYTHON)))'\' >>$@+
compat/pcre2/get-pcre2.sh new file +67 -0
67 lines not on this page · raw message
Signature · 1 line
2.11.0
PATCH 2/6 Makefile & compat/pcre2: add dependency on pcre2_convert.c
In reply to Ævar Arnfjörð BjarmasonAdd a dependency on the experimental pcre2_convert.c. This only exists in svn trunk of pcre2 currently, and allows for converting POSIX basic/extended & glob patterns to patterns accepted by PCRE[1][2].
1. https://bugs.exim.org/show_bug.cgi?id=2106
2. https://bugs.exim.org/show_bug.cgi?id=2107
--- Makefile | 1 + compat/pcre2/get-pcre2.sh | 1 + 2 files changed, 2 insertions(+)
Makefile +1 -0
@@ -1541,6 +1541,7 @@ endif compat/pcre2/src/pcre2_compile.o \ compat/pcre2/src/pcre2_config.o \ compat/pcre2/src/pcre2_context.o \ + compat/pcre2/src/pcre2_convert.o \ compat/pcre2/src/pcre2_error.o \ compat/pcre2/src/pcre2_find_bracket.o \ compat/pcre2/src/pcre2_jit_compile.o \
compat/pcre2/get-pcre2.sh +1 -0
@@ -26,6 +26,7 @@ for srcfile in \ pcre2_compile.c \ pcre2_config.c \ pcre2_context.c \ + pcre2_convert.c \ pcre2_error.c \ pcre2_find_bracket.c \ pcre2_jit_compile.c \
Signature · 1 line
2.11.0
PATCH 4/6 test-lib: add LIBPCRE1 & LIBPCRE2 prerequisites
In reply to Ævar Arnfjörð BjarmasonAdd LIBPCRE1 and LIBPCRE2 prerequisites which are true when git is compiled with USE_LIBPCRE1=YesPlease or USE_LIBPCRE2=YesPlease, respectively.
There are various edge cases or version-specific features that need to be tested for.
--- t/README | 12 ++++++++++++ t/test-lib.sh | 2 ++ 2 files changed, 14 insertions(+)
t/README +12 -0
@@ -808,6 +808,18 @@ use these, and "test_set_prereq" for how to define your own. Git was compiled with support for PCRE. Wrap any tests that use git-grep --perl-regexp or git-grep -P in these. + - LIBPCRE1 + + Git was compiled with PCRE v1 support via + USE_LIBPCRE1=YesPlease. Wrap any PCRE using tests that for some + reason need v1 of the PCRE library instead of v2 in these. + + - LIBPCRE2 + + Git was compiled with PCRE v2 support via + USE_LIBPCRE2=YesPlease. Wrap any PCRE using tests that for some + reason need v2 of the PCRE library instead of v1 in these. + - CASE_INSENSITIVE_FS Test is run on a case insensitive file system.
t/test-lib.sh +2 -0
@@ -1012,6 +1012,8 @@ test -z "$NO_PERL" && test_set_prereq PERL test -z "$NO_PTHREADS" && test_set_prereq PTHREADS test -z "$NO_PYTHON" && test_set_prereq PYTHON test -n "$USE_LIBPCRE1$USE_LIBPCRE2" && test_set_prereq PCRE +test -n "$USE_LIBPCRE1" && test_set_prereq LIBPCRE1 +test -n "$USE_LIBPCRE2" && test_set_prereq LIBPCRE2 test -z "$NO_GETTEXT" && test_set_prereq GETTEXT # Can we rely on git's output in the C locale?
Signature · 1 line
2.11.0
PATCH 5/6 grep: support regex patterns containing \0 via PCRE v2
In reply to Ævar Arnfjörð BjarmasonSupport regex patterns with embedded \0's, as an earlier commit[1] notes this was previously impossible due to an internal limitation.
Before this change any regex metacharacters in patterns containing \0 were silently ignored and the pattern matched as if it were a --fixed-strings pattern.
Now these patterns will be matched with PCRE instead, which supports combining regex metacharacters with patterns containing \0.
A side-effect of this change is that these patterns which previously would be considered --fixed-strings patterns regardless of the engine requested now all implicitly become --perl-regexp instead.
A subsequent change introduces a POSIX to PCRE syntax converter, and could be used to be 100% truthful to our documentation by using POSIX basic syntax (which we haven't been in quite some time with kwset).
But due to a chicken & egg issue with this change being easier to implement stand-alone first, the subsequent change depending on a SVN trunk version of PCRE, but most importantly I don't think anyone will mind this change, so I'm leaving it as it is.
This implementation is faster than the previous kwset implementation, but I haven't bothered to come up with a \0-specific fixed-string performance test.
See the next change in this series for a change which optionally expands the PCRE v2 use to use it for all fixed-string patterns, the performance tests for those will be applicable to these patterns as well, since PCRE matches \0 like any other character.
1. ("grep: factor test for \0 in grep patterns into a function", 2017-05-08)
--- grep.c | 24 ++++++++++++++++ t/t7008-grep-binary.sh | 74 ++++++++++++++++++++++++++++++++++---------------- 2 files changed, 75 insertions(+), 23 deletions(-)
grep.c +24 -0
@@ -613,6 +613,30 @@ static void compile_regexp(struct grep_pat *p, struct grep_opt *opt) icase = opt->regflags ®_ICASE || p->ignore_case; ascii_only = !has_non_ascii (p->pattern); +#ifdef USE_LIBPCRE2 +if (has_null (p->pattern, p->patternlen)) { +struct strbuf sb =STRBUF_INIT ; +if (icase) +strbuf_add (&sb,"(?i)" ,4 ); +if (opt->fixed) +strbuf_add (&sb,"\\Q" ,2 ); +strbuf_add (&sb, p->pattern, p->patternlen); +if (opt->fixed) +strbuf_add (&sb,"\\E" ,2 ); + + p->pattern = sb.buf; + p->patternlen = sb.len; + +/* FIXME: Check in compile_pcre2_pattern() that we're +* using basic rx using !opt->pcre2 && <something> +*/ + opt->pcre2 =1 ; + +compile_pcre2_pattern (p, opt); +return ; + } +#endif +/* * Even when -F (fixed) asks us to do a non-regexp search, we * may not be able to correctly case-fold when -i
t/t7008-grep-binary.sh +51 -23
86 lines not on this page · raw message
Signature · 1 line
2.11.0
PATCH 6/6 grep: use PCRE v2 under the hood for -G & -E for amazing speedup
In reply to Ævar Arnfjörð BjarmasonChange the underlying engine powering POSIX basic & extended patterns to be PCRE v2 under the hood.
This relies on an experimental SVN-trunk only PCRE v2 API which Philip Hazel (the PCRE maintainer) wrote up in response to a feature request I filed1[1].
This allows us to use pcre2_pattern_convert() to power all grep regex matches by converting the POSIX patterns into PCRE syntax before compiling them.
Due to PCRE generally being faster than POSIX, but most importantly due to its JIT feature (where available) this speeds up grep by a *lot*.
The improvements to the "perl" tests are already a part of this series, but all the other benchmarks show improvements made by this change alone:
$ GIT_PERF_REPEAT_COUNT=30 GIT_PERF_LARGE_REPO=~/g/linux [see note #2 for the GIT_PERF_MAKE_COMMAND ...]
[...]
Test v2.13.0 HEAD
-----------------------------------------------------------------------------------------
7810.1: grep worktree, cheap regex 0.19(0.39+0.52) 0.19(0.26+0.56) +0.0%
7810.2: grep worktree, expensive regex 5.11(29.75+0.33) 1.07(5.36+0.34) -79.1%
7810.3: grep --cached, cheap regex 2.91(2.77+0.12) 2.85(2.78+0.06) -2.1%
7810.4: grep --cached, expensive regex 21.00(20.89+0.08) 6.25(6.18+0.06) -70.2%
7820.1: basic grep how.to 0.32(1.20+0.43) 0.19(0.26+0.56) -40.6%
7820.2: extended grep how.to 0.32(1.12+0.51) 0.19(0.22+0.60) -40.6%
7820.3: perl grep how.to 0.31(1.11+0.45) 0.19(0.30+0.54) -38.7%
7820.5: basic grep ^how to 0.31(1.09+0.51) 0.19(0.24+0.57) -38.7%
7820.6: extended grep ^how to 0.31(1.14+0.46) 0.19(0.29+0.52) -38.7%
7820.7: perl grep ^how to 0.57(2.63+0.38) 0.19(0.25+0.56) -66.7%
7820.9: basic grep [how] to 0.49(2.19+0.36) 0.22(0.36+0.54) -55.1%
7820.10: extended grep [how] to 0.49(2.16+0.41) 0.22(0.41+0.50) -55.1%
7820.11: perl grep [how] to 0.57(2.55+0.40) 0.22(0.35+0.55) -61.4%
7820.13: basic grep \(e.t[^ ]*\|v.ry\) rare 0.65(3.18+0.38) 0.22(0.44+0.52) -66.2%
7820.14: extended grep (e.t[^ ]*|v.ry) rare 0.65(3.17+0.40) 0.21(0.47+0.52) -67.7%
7820.15: perl grep (e.t[^ ]*|v.ry) rare 1.05(5.64+0.34) 0.22(0.46+0.53) -79.0%
7820.17: basic grep m\(ú\|u\)ult.b\(æ\|y\)te 0.33(1.33+0.38) 0.19(0.31+0.51) -42.4%
7820.18: extended grep m(ú|u)ult.b(æ|y)te 0.33(1.27+0.44) 0.19(0.32+0.50) -42.4%
7820.19: perl grep m(ú|u)ult.b(æ|y)te 0.37(1.58+0.40) 0.19(0.30+0.53) -48.6%
7821.1: fixed grep int 0.53(1.70+0.60) 0.43(1.13+0.66) -18.9%
7821.2: basic grep int 0.55(1.62+0.59) 0.46(1.08+0.64) -16.4%
7821.3: extended grep int 0.54(1.65+0.59) 0.45(1.17+0.56) -16.7%
7821.4: perl grep int 0.54(1.63+0.62) 0.46(1.12+0.60) -14.8%
7821.6: fixed grep -i int 0.58(1.93+0.54) 0.48(1.40+0.52) -17.2%
7821.7: basic grep -i int 0.83(1.91+0.60) 0.57(1.23+0.67) -31.3%
7821.8: extended grep -i int 0.59(1.80+0.66) 0.48(1.33+0.59) -18.6%
7821.9: perl grep -i int 0.61(1.91+0.56) 0.52(1.28+0.63) -14.8%
7821.11: fixed grep æ 0.34(1.25+0.45) 0.19(0.29+0.51) -44.1%
7821.12: basic grep æ 0.34(1.26+0.43) 0.19(0.28+0.53) -44.1%
7821.13: extended grep æ 0.34(1.22+0.48) 0.19(0.29+0.53) -44.1%
7821.14: perl grep æ 0.34(1.30+0.41) 0.19(0.26+0.57) -44.1%
7821.16: fixed grep -i æ 0.27(0.88+0.46) 0.19(0.30+0.51) -29.6%
7821.17: basic grep -i æ 0.27(0.88+0.44) 0.19(0.27+0.54) -29.6%
7821.18: extended grep -i æ 0.27(0.90+0.42) 0.19(0.22+0.59) -29.6%
7821.19: perl grep -i æ 0.25(0.74+0.51) 0.18(0.27+0.58) -28.0%
7821.1: fixed grep int 0.53(1.70+0.60) 0.43(1.13+0.66) -18.9%
7821.2: basic grep int 0.55(1.62+0.59) 0.46(1.08+0.64) -16.4%
7821.3: extended grep int 0.54(1.65+0.59) 0.45(1.17+0.56) -16.7%
7821.4: perl grep int 0.54(1.63+0.62) 0.46(1.12+0.60) -14.8%
7821.6: fixed grep -i int 0.58(1.93+0.54) 0.48(1.40+0.52) -17.2%
7821.7: basic grep -i int 0.83(1.91+0.60) 0.57(1.23+0.67) -31.3%
7821.8: extended grep -i int 0.59(1.80+0.66) 0.48(1.33+0.59) -18.6%
7821.9: perl grep -i int 0.61(1.91+0.56) 0.52(1.28+0.63) -14.8%
7821.11: fixed grep æ 0.34(1.25+0.45) 0.19(0.29+0.51) -44.1%
7821.12: basic grep æ 0.34(1.26+0.43) 0.19(0.28+0.53) -44.1%
7821.13: extended grep æ 0.34(1.22+0.48) 0.19(0.29+0.53) -44.1%
7821.14: perl grep æ 0.34(1.30+0.41) 0.19(0.26+0.57) -44.1%
7821.16: fixed grep -i æ 0.27(0.88+0.46) 0.19(0.30+0.51) -29.6%
7821.17: basic grep -i æ 0.27(0.88+0.44) 0.19(0.27+0.54) -29.6%
7821.18: extended grep -i æ 0.27(0.90+0.42) 0.19(0.22+0.59) -29.6%
7821.19: perl grep -i æ 0.25(0.74+0.51) 0.18(0.27+0.58) -28.0%Caveats & other things to mention:
* This will expose PCRE v2 (as opposed to C library reg(comp|exec))to the network via gitweb in its default configuration. See <CACBZZX6V8qbnrZAdhRvPthy5Z91iEG8rrJ=Sf9tdkOt52M9j1Q@mail.gmail.com> for a discussion of security & other caveats related to that.
* I'm checking for PCRE2_CONVERT_POSIX_BASIC to enable this, but the experimental API of pcre2_pattern_convert() may change before it makes it into a release.
If we think this patch is awesome enough to get into a git release regardless, it should be guarded by some other method so we don't rudely tie upstream PCRE to this API least they break git versions in the wild.
* One way to do to that would be to guard this via the USE_LIBPCRE2_BUNDLED flag, but see the above E-Mail thread for concerns about shipping an embedded PCRE, and for ways that could be made OK.
* We could ship some copy of just the logic in pcre2_pattern_convert() & use the system PCRE instead. I haven't tried splitting it off from the PCRE codebase, and don't know how hard that would be.
* There are outstanding bugs in the pcre2_pattern_convert() function. Grepping with -G and -E for all ASCII characters from 1..127 both "$char" and "\\$char" will produce numerous differences. These are mostly obscure cases, I'm working out fixes to those with Philip.
1. https://bugs.exim.org/show_bug.cgi?id=2106
2. GIT_PERF_MAKE_COMMAND='grep -q LIBPCRE2 Makefile && make -j8 USE_LIBPCRE2=YesPlease USE_LIBPCRE2_BUNDLED=Y CC=~/perl5/installed/bin/gcc NO_R_TO_GCC_LINKER=YesPlease CFLAGS=-O3 || make -j8 USE_LIBPCRE=YesPlease CC=~/perl5/installed/bin/gcc NO_R_TO_GCC_LINKER=YesPlease CFLAGS=-O3 LIBPCREDIR=/home/avar/g/pcre/inst LDFLAGS=-Wl,-rpath,/home/avar/g/pcre/inst/lib' ./run v2.13.0 HEAD -- p*grep*
--- grep.c | 97 +++++++++++++++++++++++++++++++++++--------------- grep.h | 5 +++ t/README | 6 ++++ t/t7008-grep-binary.sh | 13 +++++-- t/test-lib.sh | 1 + 5 files changed, 90 insertions(+), 32 deletions(-)
grep.c +68 -29
@@ -472,8 +472,48 @@ static void compile_pcre2_pattern(struct grep_pat *p, const struct grep_opt *optconst uint8_t *character_tables =NULL ;uint32_t canjit;int jitret; +int icase = opt->regflags ®_ICASE || p->ignore_case; +PCRE2_SPTR pattern = (PCRE2_SPTR )p->pattern; +PCRE2_SIZE length = p->patternlen; +int copied_pattern =0 ; +struct strbuf pattern_sb =STRBUF_INIT ; +#ifdef PCRE2_CONVERT_POSIX_BASIC +int convret; +PCRE2_UCHAR *convpatbuf =NULL ; +PCRE2_SIZE convpatlen; +int converted_pattern =0 ; +#endif -assert (opt->pcre2); +if (opt->fixed ||has_null (p->pattern, p->patternlen) ||is_fixed (p->pattern, p->patternlen)) { +if (icase) +strbuf_add (&pattern_sb,"(?i)" ,4 ); +if (opt->fixed) +strbuf_add (&pattern_sb,"\\Q" ,2 ); +strbuf_add (&pattern_sb, p->pattern, p->patternlen); +if (opt->fixed) +strbuf_add (&pattern_sb,"\\E" ,2 ); + + pattern = (PCRE2_SPTR )pattern_sb.buf; + length = pattern_sb.len; + copied_pattern =1 ; + }else if (opt->pcre2_posix_emulation) { +#ifdef PCRE2_CONVERT_POSIX_BASIC + convret =pcre2_pattern_convert (pattern, length, + (opt->regflags ®_EXTENDED + ?PCRE2_CONVERT_POSIX_EXTENDED + :PCRE2_CONVERT_POSIX_BASIC ), + &convpatbuf, &convpatlen,NULL ); +if (convret !=0 ) { +fprintf (stderr,"oh noes\n" ); +pcre2_get_error_message (convret, errbuf,sizeof (errbuf)); +compile_regexp_failed (p, (const char *)&errbuf); + } + pattern = convpatbuf; + length = convpatlen; + converted_pattern =1 ; +#endif + }else +assert (opt->pcre2); p->pcre2_compile_context =NULL ;
@@ -488,11 +528,16 @@ static void compile_pcre2_pattern(struct grep_pat *p, const struct grep_opt *optif (is_utf8_locale () &&has_non_ascii (p->pattern)) options |=PCRE2_UTF ; - p->pcre2_pattern =pcre2_compile ((PCRE2_SPTR )p->pattern, - p->patternlen, options, &error, &erroffset, - p->pcre2_compile_context); + p->pcre2_pattern =pcre2_compile (pattern, length, options, &error, + &erroffset, p->pcre2_compile_context);if (p->pcre2_pattern) { +if (copied_pattern) +strbuf_release (&pattern_sb); +#ifdef PCRE2_CONVERT_POSIX_BASIC +if (converted_pattern) +pcre2_converted_pattern_free (convpatbuf); +#endif p->pcre2_match_data =pcre2_match_data_create_from_pattern (p->pcre2_pattern,NULL );if (!p->pcre2_match_data)die ("BUG: Couldn't allocate PCRE2 match data" );
@@ -580,7 +625,6 @@ static int pcre2match(struct grep_pat *p, const char *line, const char *eol,static void free_pcre2_pattern (struct grep_pat *p) { } -#endif /* !USE_LIBPCRE2 */ static void compile_fixed_regexp (struct grep_pat *p,struct grep_opt *opt) {
@@ -602,41 +646,21 @@ static void compile_fixed_regexp(struct grep_pat *p, struct grep_opt *opt) compile_regexp_failed(p, errbuf); } } +#endif /* !USE_LIBPCRE2 */ static void compile_regexp (struct grep_pat *p,struct grep_opt *opt) { +#ifndef USE_LIBPCRE2 int icase, ascii_only; +#endif int err; p->word_regexp = opt->word_regexp; p->ignore_case = opt->ignore_case; +#ifndef USE_LIBPCRE2 icase = opt->regflags ®_ICASE || p->ignore_case; ascii_only = !has_non_ascii (p->pattern); -#ifdef USE_LIBPCRE2 -if (has_null (p->pattern, p->patternlen)) { -struct strbuf sb =STRBUF_INIT ; -if (icase) -strbuf_add (&sb,"(?i)" ,4 ); -if (opt->fixed) -strbuf_add (&sb,"\\Q" ,2 ); -strbuf_add (&sb, p->pattern, p->patternlen); -if (opt->fixed) -strbuf_add (&sb,"\\E" ,2 ); - - p->pattern = sb.buf; - p->patternlen = sb.len; - -/* FIXME: Check in compile_pcre2_pattern() that we're -* using basic rx using !opt->pcre2 && <something> -*/ - opt->pcre2 =1 ; - -compile_pcre2_pattern (p, opt); -return ; - } -#endif -/* * Even when -F (fixed) asks us to do a non-regexp search, we * may not be able to correctly case-fold when -i
@@ -668,12 +692,26 @@ static void compile_regexp(struct grep_pat *p, struct grep_opt *opt)compile_fixed_regexp (p, opt); return; } +#endif if (opt->pcre2) {compile_pcre2_pattern (p, opt);return ; } +#ifdef USE_LIBPCRE2 +if (opt->fixed ||has_null (p->pattern, p->patternlen) ||is_fixed (p->pattern, p->patternlen)) { +compile_pcre2_pattern (p, opt); +return ; + } + +#ifdef PCRE2_CONVERT_POSIX_BASIC + opt->pcre2_posix_emulation =1 ; +compile_pcre2_pattern (p, opt); +return ; +#endif +#endif +if (opt->pcre1) {compile_pcre1_regexp (p, opt);return ;
@@ -686,6 +724,7 @@ static void compile_regexp(struct grep_pat *p, struct grep_opt *opt)regfree (&p->regexp);compile_regexp_failed (p, errbuf); } +return ; }static struct grep_expr *compile_pattern_or (struct grep_pat **);
grep.h +5 -0
@@ -29,6 +29,9 @@ typedef int pcre2_compile_context;typedef int pcre2_match_context ;typedef int pcre2_jit_stack ;#endif +#ifndef PCRE2_CONVERT_POSIX_EXTENDED +typedef int pcre2_convert_context ; +#endif #include "kwset.h" #include "thread-utils.h" #include "userdiff.h"
@@ -73,6 +76,7 @@ struct grep_pat {pcre_jit_stack *pcre1_jit_stack;const unsigned char *pcre1_tables;int pcre1_jit_on; +pcre2_convert_context *pcre2_convert_context;pcre2_code *pcre2_pattern;pcre2_match_data *pcre2_match_data;pcre2_compile_context *pcre2_compile_context;
@@ -143,6 +147,7 @@ struct grep_opt {int use_reflog_filter;int pcre1;int pcre2; +int pcre2_posix_emulation;int relative;int pathname;int null_following_name;
t/README +6 -0
@@ -820,6 +820,12 @@ use these, and "test_set_prereq" for how to define your own. USE_LIBPCRE2=YesPlease. Wrap any PCRE using tests that for some reason need v2 of the PCRE library instead of v1 in these. + - LIBPCRE2_BUNDLED + + Git was compiled with the bundled PCRE v2 support via + USE_LIBPCRE2=YesPlease & + USE_LIBPCRE2_BUNDLED=IWantPatternConvertAwesomeSauce. + - CASE_INSENSITIVE_FS Test is run on a case insensitive file system.
t/t7008-grep-binary.sh +10 -3
@@ -100,9 +100,16 @@ test_expect_success 'git grep ile a' ' git grep ile a ' -test_expect_failure 'git grep .fi a' ' - git grep .fi a -' +if test_have_prereq LIBPCRE2_BUNDLED +then + test_expect_success 'git grep .fi a' ' + git grep .fi a + ' +else + test_expect_failure 'git grep .fi a' ' + git grep .fi a + ' +fi nul_match 1 '-F' 'yQf' nul_match 0 '-F' 'yQx'
t/test-lib.sh +1 -0
@@ -1014,6 +1014,7 @@ test -z "$NO_PYTHON" && test_set_prereq PYTHON test -n "$USE_LIBPCRE1$USE_LIBPCRE2" && test_set_prereq PCRE test -n "$USE_LIBPCRE1" && test_set_prereq LIBPCRE1 test -n "$USE_LIBPCRE2" && test_set_prereq LIBPCRE2 +test -n "$USE_LIBPCRE2_BUNDLED" && test_set_prereq LIBPCRE2_BUNDLED test -z "$NO_GETTEXT" && test_set_prereq GETTEXT # Can we rely on git's output in the C locale?
Signature · 1 line
2.11.0