[PATCH v3] grep: fix multibyte regex handling under macOS

[PATCH v3] grep: fix multibyte regex handling under macOS

[Diomidis Spinellis]

The commit 29de20504e (Makefile: fix default regex settings on
Darwin, 2013-05-11) fixed t0070-fundamental.sh under Darwin (macOS) by
adopting Git's regex library.  However, this library is compiled with
NO_MBSUPPORT, which causes git-grep to work incorrectly on multibyte
(e.g. UTF-8) files.  Current macOS versions pass t0070-fundamental.sh
with the native macOS regex library, which also supports multibyte
characters.

Adjust the Makefile to use the native regex library, and call
setlocale(3) to set CTYPE according to the user's preference.
The setlocale call is required on all platforms, but in platforms
supporting gettext(3), setlocale was called as a side-effect of
initializing gettext.  Therefore, move the CTYPE setlocale call from
gettext.c to common-main.c and the corresponding locale.h include
into git-compat-util.h.

To avoid running the new tests on the few platforms still using the
compatibility library, which is compiled without multibyte support,
store the corresponding NO_REGEX setting in the GIT-BUILD-OPTIONS file.
This makes it available to the test scripts.  Thanks to the global
initialization of CTYPE setlocale, the test-tool regex command now
works correctly with multibyte regexes and is used to further test
a platform's support for them.

Signed-off-by: Diomidis Spinellis <dds@aueb.gr>
---
- Move added setlocale from grep.c and test-regex.c to common-main.c
- Move added locale.h include from the above files to git-compat-util.h
- Correct use of test_expect_code 1
- Have old commit reference in the commit message follow commit
  referencing conventions

 Makefile                  |  2 +-
 common-main.c             |  1 +
 gettext.c                 |  2 --
 git-compat-util.h         |  1 +
 t/t7818-grep-multibyte.sh | 34 ++++++++++++++++++++++++++++++++++
 5 files changed, 37 insertions(+), 3 deletions(-)
 create mode 100755 t/t7818-grep-multibyte.sh

diff --git a/Makefile b/Makefile
index 04d0fd1fe6..d1a9825715 100644
--- a/Makefile
+++ b/Makefile
@@ -1427,7 +1427,6 @@ ifeq ($(uname_S),Darwin)
 		APPLE_COMMON_CRYPTO = YesPlease
 		COMPAT_CFLAGS += -DAPPLE_COMMON_CRYPTO
 	endif
-	NO_REGEX = YesPlease
 	PTHREAD_LIBS =
 endif
 
@@ -2970,6 +2969,7 @@ GIT-BUILD-OPTIONS: FORCE
 	@echo NO_PERL=\''$(subst ','\'',$(subst ','\'',$(NO_PERL)))'\' >>$@+
 	@echo NO_PTHREADS=\''$(subst ','\'',$(subst ','\'',$(NO_PTHREADS)))'\' >>$@+
 	@echo NO_PYTHON=\''$(subst ','\'',$(subst ','\'',$(NO_PYTHON)))'\' >>$@+
+	@echo NO_REGEX=\''$(subst ','\'',$(subst ','\'',$(NO_REGEX)))'\' >>$@+
 	@echo NO_UNIX_SOCKETS=\''$(subst ','\'',$(subst ','\'',$(NO_UNIX_SOCKETS)))'\' >>$@+
 	@echo PAGER_ENV=\''$(subst ','\'',$(subst ','\'',$(PAGER_ENV)))'\' >>$@+
 	@echo DC_SHA1=\''$(subst ','\'',$(subst ','\'',$(DC_SHA1)))'\' >>$@+
diff --git a/common-main.c b/common-main.c
index c531372f3f..0a22861f1c 100644
--- a/common-main.c
+++ b/common-main.c
@@ -40,6 +40,7 @@ int main(int argc, const char **argv)
 
 	git_resolve_executable_dir(argv[0]);
 
+	setlocale(LC_CTYPE, "");
 	git_setup_gettext();
 
 	initialize_the_repository();
diff --git a/gettext.c b/gettext.c
index bb5ba1fe7c..f139008d0a 100644
--- a/gettext.c
+++ b/gettext.c
@@ -10,7 +10,6 @@
 #include "config.h"
 
 #ifndef NO_GETTEXT
-#	include <locale.h>
 #	include <libintl.h>
 #	ifdef GIT_WINDOWS_NATIVE
 
@@ -80,7 +79,6 @@ static int test_vsnprintf(const char *fmt, ...)
 
 static void init_gettext_charset(const char *domain)
 {
-	setlocale(LC_CTYPE, "");
 	charset = locale_charset();
 	bind_textdomain_codeset(domain, charset);
 
diff --git a/git-compat-util.h b/git-compat-util.h
index 58d7708296..c6fa3c7469 100644
--- a/git-compat-util.h
+++ b/git-compat-util.h
@@ -212,6 +212,7 @@
 #endif
 #include <errno.h>
 #include <limits.h>
+#include <locale.h>
 #ifdef NEEDS_SYS_PARAM_H
 #include <sys/param.h>
 #endif
diff --git a/t/t7818-grep-multibyte.sh b/t/t7818-grep-multibyte.sh
new file mode 100755
index 0000000000..a3889f9822
--- /dev/null
+++ b/t/t7818-grep-multibyte.sh
@@ -0,0 +1,34 @@
+#!/bin/sh
+
+test_description='grep multibyte characters'
+
+. ./test-lib.sh
+
+# Multibyte regex search is only supported with a native regex library
+# that supports it.
+# (The supplied compatibility library is compiled with NO_MBSUPPORT.)
+test -z "$NO_REGEX" &&
+  LC_ALL=en_US.UTF-8 test-tool regex '^.$' '¿' &&
+  test_set_prereq MB_REGEX
+
+if ! test_have_prereq MB_REGEX
+then
+  skip_all='multibyte grep tests; Git compiled with NO_REGEX, NO_MBSUPPORT'
+  test_done
+fi
+
+test_expect_success 'setup' '
+	test_write_lines "¿" >file &&
+	git add file &&
+	LC_ALL="en_US.UTF-8" &&
+	export LC_ALL
+'
+test_expect_success 'grep exactly one char in single-char multibyte file' '
+	git grep "^.$"
+'
+
+test_expect_success 'grep two chars in single-char multibyte file' '
+	test_expect_code 1 git grep ".."
+'
+
+test_done
-- 
2.30.2

Re: [PATCH v3] grep: fix multibyte regex handling under macOS

[Junio C Hamano]

Diomidis Spinellis <dds@aueb.gr> writes:

> diff --git a/Makefile b/Makefile
> index 04d0fd1fe6..d1a9825715 100644
> --- a/Makefile
> +++ b/Makefile
> @@ -1427,7 +1427,6 @@ ifeq ($(uname_S),Darwin)
>  		APPLE_COMMON_CRYPTO = YesPlease
>  		COMPAT_CFLAGS += -DAPPLE_COMMON_CRYPTO
>  	endif
> -	NO_REGEX = YesPlease
>  	PTHREAD_LIBS =
>  endif
>  
> @@ -2970,6 +2969,7 @@ GIT-BUILD-OPTIONS: FORCE
>  	@echo NO_PERL=\''$(subst ','\'',$(subst ','\'',$(NO_PERL)))'\' >>$@+
>  	@echo NO_PTHREADS=\''$(subst ','\'',$(subst ','\'',$(NO_PTHREADS)))'\' >>$@+
>  	@echo NO_PYTHON=\''$(subst ','\'',$(subst ','\'',$(NO_PYTHON)))'\' >>$@+
> +	@echo NO_REGEX=\''$(subst ','\'',$(subst ','\'',$(NO_REGEX)))'\' >>$@+
>  	@echo NO_UNIX_SOCKETS=\''$(subst ','\'',$(subst ','\'',$(NO_UNIX_SOCKETS)))'\' >>$@+
>  	@echo PAGER_ENV=\''$(subst ','\'',$(subst ','\'',$(PAGER_ENV)))'\' >>$@+
>  	@echo DC_SHA1=\''$(subst ','\'',$(subst ','\'',$(DC_SHA1)))'\' >>$@+

Build part looks good to me.

> diff --git a/common-main.c b/common-main.c
> index c531372f3f..0a22861f1c 100644
> --- a/common-main.c
> +++ b/common-main.c
> @@ -40,6 +40,7 @@ int main(int argc, const char **argv)
>  
>  	git_resolve_executable_dir(argv[0]);
>  
> +	setlocale(LC_CTYPE, "");
>  	git_setup_gettext();
>  
>  	initialize_the_repository();
> diff --git a/gettext.c b/gettext.c
> index bb5ba1fe7c..f139008d0a 100644
> --- a/gettext.c
> +++ b/gettext.c
> @@ -10,7 +10,6 @@
>  #include "config.h"
>  
>  #ifndef NO_GETTEXT
> -#	include <locale.h>
>  #	include <libintl.h>
>  #	ifdef GIT_WINDOWS_NATIVE
>  
> @@ -80,7 +79,6 @@ static int test_vsnprintf(const char *fmt, ...)
>  
>  static void init_gettext_charset(const char *domain)
>  {
> -	setlocale(LC_CTYPE, "");
>  	charset = locale_charset();
>  	bind_textdomain_codeset(domain, charset);
> 
> diff --git a/git-compat-util.h b/git-compat-util.h
> index 58d7708296..c6fa3c7469 100644
> --- a/git-compat-util.h
> +++ b/git-compat-util.h
> @@ -212,6 +212,7 @@
>  #endif
>  #include <errno.h>
>  #include <limits.h>
> +#include <locale.h>
>  #ifdef NEEDS_SYS_PARAM_H
>  #include <sys/param.h>
>  #endif

I'll let others more familiar with the locale support to comment on
these changes.  We are unconditionally including <locale.h> now;
before platforms that lack locale.h can set NO_GETTEXT but that will
no longer work as a "workaround" for them.  I do not know if thta is
a practical downside to anybody, but it could be a problem.

> diff --git a/t/t7818-grep-multibyte.sh b/t/t7818-grep-multibyte.sh
> new file mode 100755
> index 0000000000..a3889f9822
> --- /dev/null
> +++ b/t/t7818-grep-multibyte.sh

Do we need a new test script for this?

> @@ -0,0 +1,34 @@
> +#!/bin/sh
> +
> +test_description='grep multibyte characters'
> +
> +. ./test-lib.sh
> +
> +# Multibyte regex search is only supported with a native regex library
> +# that supports it.
> +# (The supplied compatibility library is compiled with NO_MBSUPPORT.)

This file is not specific to Darwin; "... with a native regex
library" etc. is not something we want to see here.

> +test -z "$NO_REGEX" &&
> +  LC_ALL=en_US.UTF-8 test-tool regex '^.$' '¿' &&
> +  test_set_prereq MB_REGEX

We can safely drop 'test -z "$NO_REGEX" &&' part here, no?  Even if
we omit it, those who built with NO_REGEX would fail "test-tool
regex" step above.  And by omitting $NO_REGEX check, we do not have
to look for and update the condition when the fallback regex engine
we use starts supporting MB_REGEX.

> +if ! test_have_prereq MB_REGEX
> +then
> +  skip_all='multibyte grep tests; Git compiled with NO_REGEX, NO_MBSUPPORT'
> +  test_done
> +fi

I do not think if we need a single use prereq here.  We can just
use whatever condition that is used to set MB_REGEX above and do the
skip-all thing here.

> +test_expect_success 'setup' '
> +	test_write_lines "¿" >file &&
> +	git add file &&
> +	LC_ALL="en_US.UTF-8" &&
> +	export LC_ALL
> +'

Missing inter-test blank line.

> +test_expect_success 'grep exactly one char in single-char multibyte file' '
> +	git grep "^.$"
> +'

I am not sure how much value we are getting out of this test, which
is identical to what we already tested earlier above with "test-tool
regex".

> +test_expect_success 'grep two chars in single-char multibyte file' '
> +	test_expect_code 1 git grep ".."
> +'
> +
> +test_done

[PATCH v4] grep: fix multibyte regex handling under macOS

[Diomidis Spinellis]

The commit 29de20504e (Makefile: fix default regex settings on
Darwin, 2013-05-11) fixed t0070-fundamental.sh under Darwin (macOS) by
adopting Git's regex library.  However, this library is compiled with
NO_MBSUPPORT, which causes git-grep to work incorrectly on multibyte
(e.g. UTF-8) files.  Current macOS versions pass t0070-fundamental.sh
with the native macOS regex library, which also supports multibyte
characters.

Adjust the Makefile to use the native regex library, and call
setlocale(3) to set CTYPE according to the user's preference.
The setlocale call is required on all platforms, but in platforms
supporting gettext(3), setlocale was called as a side-effect of
initializing gettext.  Therefore, move the CTYPE setlocale call from
gettext.c to common-main.c and the corresponding locale.h include
into git-compat-util.h.

Thanks to the global initialization of CTYPE setlocale, the test-tool
regex command now works correctly with supported multibyte regexes, and
is used to set the MB_REGEX test prerequisite by assessing a platform's
support for them.

Signed-off-by: Diomidis Spinellis <dds@aueb.gr>
---

- Fold test script into t7810-grep.sh
- Simplify the test prerequisite evaluation

 Makefile          |  2 +-
 common-main.c     |  1 +
 gettext.c         |  2 --
 git-compat-util.h |  1 +
 t/t7810-grep.sh   | 15 +++++++++++++++
 5 files changed, 18 insertions(+), 3 deletions(-)

diff --git a/Makefile b/Makefile
index 04d0fd1fe6..d1a9825715 100644
--- a/Makefile
+++ b/Makefile
@@ -1427,7 +1427,6 @@ ifeq ($(uname_S),Darwin)
 		APPLE_COMMON_CRYPTO = YesPlease
 		COMPAT_CFLAGS += -DAPPLE_COMMON_CRYPTO
 	endif
-	NO_REGEX = YesPlease
 	PTHREAD_LIBS =
 endif
 
@@ -2970,6 +2969,7 @@ GIT-BUILD-OPTIONS: FORCE
 	@echo NO_PERL=\''$(subst ','\'',$(subst ','\'',$(NO_PERL)))'\' >>$@+
 	@echo NO_PTHREADS=\''$(subst ','\'',$(subst ','\'',$(NO_PTHREADS)))'\' >>$@+
 	@echo NO_PYTHON=\''$(subst ','\'',$(subst ','\'',$(NO_PYTHON)))'\' >>$@+
+	@echo NO_REGEX=\''$(subst ','\'',$(subst ','\'',$(NO_REGEX)))'\' >>$@+
 	@echo NO_UNIX_SOCKETS=\''$(subst ','\'',$(subst ','\'',$(NO_UNIX_SOCKETS)))'\' >>$@+
 	@echo PAGER_ENV=\''$(subst ','\'',$(subst ','\'',$(PAGER_ENV)))'\' >>$@+
 	@echo DC_SHA1=\''$(subst ','\'',$(subst ','\'',$(DC_SHA1)))'\' >>$@+
diff --git a/common-main.c b/common-main.c
index c531372f3f..0a22861f1c 100644
--- a/common-main.c
+++ b/common-main.c
@@ -40,6 +40,7 @@ int main(int argc, const char **argv)
 
 	git_resolve_executable_dir(argv[0]);
 
+	setlocale(LC_CTYPE, "");
 	git_setup_gettext();
 
 	initialize_the_repository();
diff --git a/gettext.c b/gettext.c
index bb5ba1fe7c..f139008d0a 100644
--- a/gettext.c
+++ b/gettext.c
@@ -10,7 +10,6 @@
 #include "config.h"
 
 #ifndef NO_GETTEXT
-#	include <locale.h>
 #	include <libintl.h>
 #	ifdef GIT_WINDOWS_NATIVE
 
@@ -80,7 +79,6 @@ static int test_vsnprintf(const char *fmt, ...)
 
 static void init_gettext_charset(const char *domain)
 {
-	setlocale(LC_CTYPE, "");
 	charset = locale_charset();
 	bind_textdomain_codeset(domain, charset);
 
diff --git a/git-compat-util.h b/git-compat-util.h
index 58d7708296..c6fa3c7469 100644
--- a/git-compat-util.h
+++ b/git-compat-util.h
@@ -212,6 +212,7 @@
 #endif
 #include <errno.h>
 #include <limits.h>
+#include <locale.h>
 #ifdef NEEDS_SYS_PARAM_H
 #include <sys/param.h>
 #endif
diff --git a/t/t7810-grep.sh b/t/t7810-grep.sh
index 6935601171..308a3dc231 100755
--- a/t/t7810-grep.sh
+++ b/t/t7810-grep.sh
@@ -18,6 +18,9 @@ test_invalid_grep_expression() {
 	'
 }
 
+LC_ALL=en_US.UTF-8 test-tool regex '^.$' '¿' &&
+  test_set_prereq MB_REGEX
+
 cat >hello.c <<EOF
 #include <assert.h>
 #include <stdio.h>
@@ -87,6 +90,10 @@ test_expect_success setup '
 		echo unusual >"\"unusual\" pathname" &&
 		echo unusual >"t/nested \"unusual\" pathname"
 	fi &&
+	if test_have_prereq MB_REGEX
+	then
+		echo "¿" >reverse-question-mark
+	fi &&
 	git add . &&
 	test_tick &&
 	git commit -m initial
@@ -568,6 +575,14 @@ do
 	'
 done
 
+test_expect_success MB_REGEX 'grep exactly one char in single-char multibyte file' '
+	LC_ALL=en_US.UTF-8 git grep "^.$" reverse-question-mark
+'
+
+test_expect_success MB_REGEX 'grep two chars in single-char multibyte file' '
+	LC_ALL=en_US.UTF-8 test_expect_code 1 git grep ".." reverse-question-mark
+'
+
 cat >expected <<EOF
 file
 EOF
-- 
2.30.2

Re: [PATCH v4] grep: fix multibyte regex handling under macOS

[Junio C Hamano]

Diomidis Spinellis <dds@aueb.gr> writes:

> - Fold test script into t7810-grep.sh
> - Simplify the test prerequisite evaluation

The tests certainly look better, and as v3, the build procedure
sounds OK.

> diff --git a/common-main.c b/common-main.c
> index c531372f3f..0a22861f1c 100644
> --- a/common-main.c
> +++ b/common-main.c
> @@ -40,6 +40,7 @@ int main(int argc, const char **argv)
>  
>  	git_resolve_executable_dir(argv[0]);
>  
> +	setlocale(LC_CTYPE, "");
>  	git_setup_gettext();
>  
>  	initialize_the_repository();
> ...
> diff --git a/git-compat-util.h b/git-compat-util.h
> index 58d7708296..c6fa3c7469 100644
> --- a/git-compat-util.h
> +++ b/git-compat-util.h
> @@ -212,6 +212,7 @@
>  #endif
>  #include <errno.h>
>  #include <limits.h>
> +#include <locale.h>
>  #ifdef NEEDS_SYS_PARAM_H
>  #include <sys/param.h>
>  #endif

I'll let others more familiar with the locale support to comment on
these changes.  We are unconditionally including <locale.h> now;
platforms that lack <locale.h> could set NO_GETTEXT to work it
around before this change, but that will no longer work.

I do not know if thta is a practical downside to anybody, but it
could be a problem.

Perhaps cook this in 'next' and see if anybody screams?

Thanks.

Re: [PATCH v4] grep: fix multibyte regex handling under macOS

[Diomidis Spinellis]

On 13-Sep-22 20:32, Junio C Hamano wrote:
>> +#include <locale.h>
>>   #ifdef NEEDS_SYS_PARAM_H
>>   #include <sys/param.h>
>>   #endif
> 
> I'll let others more familiar with the locale support to comment on
> these changes.  We are unconditionally including <locale.h> now;
> platforms that lack <locale.h> could set NO_GETTEXT to work it
> around before this change, but that will no longer work.
> 
> I do not know if thta is a practical downside to anybody, but it
> could be a problem.
> 
> Perhaps cook this in 'next' and see if anybody screams?

I was programming in C before locale.h was introduced, so its 
availability was something that worried me as well.  I looked up whether 
we could/should use a GNU autoconf macro to test for the locale.h 
availability.  According to autoconf's documentation "All hosted 
environments that are still of interest for portable code provide all of 
the headers specified in ISO C90 (as amended in 1995): assert.h, [...] 
locale.h [...], and wctype.h. Most programs can safely include these 
headers unconditionally." [1]

[1] https://www.gnu.org/software/autoconf/manual/autoconf-2.70/autoconf.html

Re: [PATCH v4] grep: fix multibyte regex handling under macOS

[Junio C Hamano]

Diomidis Spinellis <dds@aueb.gr> writes:

>> Perhaps cook this in 'next' and see if anybody screams?

Let's do this.

> I was programming in C before locale.h was introduced, so its
> availability was something that worried me as well.  I looked up
> whether we could/should use a GNU autoconf macro to test for the
> locale.h availability.  According to autoconf's documentation "All
> hosted environments that are still of interest for portable code
> provide all of the headers specified in ISO C90 (as amended in 1995):
> assert.h, [...] locale.h [...], and wctype.h. Most programs can safely
> include these headers unconditionally." [1]

Yeah, but there are folks on minor platforms where full POSIX
compliance is iffy, so whatever autoconf's documentation says
unfortunately does not weigh as much as we wish it does.