portability: allow building in systems without d_type
PATCH 1/1 portability: allow building in systems without d_type
Since 09fb155f11 (diff --no-index: support limiting by pathspec, 2025-05-21) will fail to build in platforms that don't have a d_type member on their struct dirent (ex: AIX, NonStop).
Use the DTYPE() macro instead of a nake reference to d_type.
--- diff-no-index.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
diff-no-index.c +1 -1
@@ -46,7 +46,7 @@ static int read_directory_contents(const char *path, struct string_list *list,if (!match_leading_pathspec (NULL , pathspec, match.buf, match.len, -0 ,NULL , e->d_type ==DT_DIR ?1 :0 )) +0 ,NULL ,DTYPE (e) ==DT_DIR ?1 :0 ))continue ; }
Signature · 1 line
2.50.0.53.g63c9ac04f7
Hi Carlo,
Quoted 16 lines · Carlo Marcelo Arenas Belón
Since
09fb155f11(diff --no-index: support limiting by pathspec, 2025-05-21) will fail to build in platforms that don't have ad_typemember on their struct dirent (ex: AIX, NonStop).Use the
DTYPE()macro instead of a nake reference tod_type.Signed-off-by Carlo Marcelo Arenas Belón <[email protected]>--- diff-no-index.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)diff-no-index.c +0
-0
diff-no-index.c static int read_directory_contents(const char *path, struct string_list *list, · line 51
Quoted 7 lines
diff-no-index.c +1
-1@@ -46,7 +46,7 @@ static int read_directory_contents(const char *path, struct string_list *list,if (!match_leading_pathspec (NULL , pathspec, match.buf, match.len, -0 ,NULL , e->d_type ==DT_DIR ?1 :0 )) +0 ,NULL ,DTYPE (e) ==DT_DIR ?1 :0 ))continue ; }
I confirm that before this patch the build will fail with the following on AIX 7.3:
CC diff-no-index.o
diff-no-index.c: In function 'read_directory_contents':
diff-no-index.c:49:21: error: 'struct dirent' has no member named 'd_type'
49 | 0, NULL, e->d_type == DT_DIR ? 1 : 0))
| ^~
gmake: *** [Makefile:2821: diff-no-index.o] Error 1
This patch fixes it. Thanks.
Collin
Quoted 2 lines · Carlo Marcelo Arenas Belón
Since
09fb155f11(diff --no-index: support limiting by pathspec, 2025-05-21) will fail to build in platforms that don't have a
s/will fail/git fails/
Quoted 3 lines
d_typemember on their struct dirent (ex: AIX, NonStop).Use the
DTYPE()macro instead of a nake reference tod_type.
s/nake/naked/
M.
Quoted 18 lines
Signed-off-by Carlo Marcelo Arenas Belón <[email protected]>--- diff-no-index.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)diff-no-index.c +1
-1@@ -46,7 +46,7 @@ static int read_directory_contents(const char *path, struct string_list *list,if (!match_leading_pathspec (NULL , pathspec, match.buf, match.len, -0 ,NULL , e->d_type ==DT_DIR ?1 :0 )) +0 ,NULL ,DTYPE (e) ==DT_DIR ?1 :0 ))continue ; }
Quoted 3 lines · Carlo Marcelo Arenas Belón
Since
09fb155f11(diff --no-index: support limiting by pathspec, 2025-05-21) will fail to build in platforms that don't have ad_typemember on their struct dirent (ex: AIX, NonStop).
s/build in/build on/
Signature · 1 line
Kristoffer Haugsbakk
Quoted 5 lines · Carlo Marcelo Arenas Belón
Since
09fb155f11(diff --no-index: support limiting by pathspec, 2025-05-21) will fail to build in platforms that don't have ad_typemember on their struct dirent (ex: AIX, NonStop).Use the
DTYPE()macro instead of a nake reference tod_type.
This may allow you to compile and build, but does the resulting binary do what you want it to?
Review
Quoted 4 lines
(quoted diff) +1
-1if (!match_leading_pathspec(NULL, pathspec, match.buf, match.len, - 0, NULL, e->d_type == DT_DIR ? 1 : 0)) + 0, NULL, DTYPE(e) == DT_DIR ? 1 : 0))
On a platform without d_type member, DTYPE() macro gives DT_UNKNOWN that is not DT_DIR, so essentially you are always passing 0 even when you are looking at a directory (in which case you must pass 1) to match_leading_pathspec().
So I somehow doubt this is a correct fix.
I do not know if get_dtype() helper function is easily applicable to this codepath, so I wrote this in a longhand...
diff-no-index.c | 18 +++++++++++++++++- 1 file changed, 17 insertions(+), 1 deletion(-)
diff-no-index.c +17 -1
@@ -41,12 +41,28 @@ static int read_directory_contents(const char *path, struct string_list *list,while ((e =readdir_skip_dot_and_dotdot (dir))) {if (pathspec) { +int is_dir =0 ; +strbuf_setlen (&match, len);strbuf_addstr (&match, e->d_name); +if (dtype !=DT_UNKNOWN ) { + is_dir = dtype ==DT_DIR ; + }else { +struct stat st; +struct strbuf pathbuf =STRBUF_INIT ; +strbuf_addstr (&pathbuf, path); +strbuf_complete (&pathbuf,'/' ); +strbuf_addstr (&pathbuf, e->d_name); +if (!lstat (&st, pathbuf.buf)) + is_dir =S_ISDIR (st.st_mode); +else + ;/* punt */ +strbuf_release (&pathbuf); + }if (!match_leading_pathspec (NULL , pathspec, match.buf, match.len, -0 ,NULL ,DTYPE (e) ==DT_DIR ?1 :0 )) +0 ,NULL , is_dir))continue ; }
Hi Junio,
Quoted 9 lines · Junio C Hamano
This may allow you to compile and build, but does the resulting binary do what you want it to?
[...]
On a platform withoutd_typemember,DTYPE()macro givesDT_UNKNOWNthat is notDT_DIR, so essentially you are always passing 0 even when you are looking at a directory (in which case you must pass 1) tomatch_leading_pathspec().So I somehow doubt this is a correct fix.
Good points. I guess I had reviewed the original patch too late to realize myself...
Quoted 31 lines
I do not know if
get_dtype()helper function is easily applicable to this codepath, so I wrote this in a longhand...diff-no-index.c | 18 +++++++++++++++++- 1 file changed, 17 insertions(+), 1 deletion(-)diff-no-index.c +14
-0@@ -41,12 +41,28 @@ static int read_directory_contents(const char *path, struct string_list *list,while ((e =readdir_skip_dot_and_dotdot (dir))) {if (pathspec) { +int is_dir =0 ; +strbuf_setlen (&match, len);strbuf_addstr (&match, e->d_name); +if (dtype !=DT_UNKNOWN ) { + is_dir = dtype ==DT_DIR ; + }else { +struct stat st; +struct strbuf pathbuf =STRBUF_INIT ; +strbuf_addstr (&pathbuf, path); +strbuf_complete (&pathbuf,'/' ); +strbuf_addstr (&pathbuf, e->d_name); +if (!lstat (&st, pathbuf.buf)) + is_dir =S_ISDIR (st.st_mode); +else + ;/* punt */
diff-no-index.c static int read_directory_contents(const char *path, struct string_list *list, · line 68
Quoted 10 lines
diff-no-index.c +3
-1+strbuf_release (&pathbuf); + }if (!match_leading_pathspec (NULL , pathspec, match.buf, match.len, -0 ,NULL ,DTYPE (e) ==DT_DIR ?1 :0 )) +0 ,NULL , is_dir))continue ; }
Two very minor issues with with this patch. The arguments to 'lstat' are reversed and the 'dtype' variable is not declared. Here is a diff that I applied after your diff:
diff-no-index.c +2 -1
@@ -42,6 +42,7 @@ static int read_directory_contents(const char *path, struct string_list *list,while ((e =readdir_skip_dot_and_dotdot (dir))) {if (pathspec) {int is_dir =0 ; +int dtype =DTYPE (e);strbuf_setlen (&match, len);strbuf_addstr (&match, e->d_name);
@@ -53,7 +54,7 @@ static int read_directory_contents(const char *path, struct string_list *list,strbuf_addstr (&pathbuf, path);strbuf_complete (&pathbuf,'/' );strbuf_addstr (&pathbuf, e->d_name); -if (!lstat (&st, pathbuf.buf)) +if (!lstat (pathbuf.buf, &st)) is_dir =S_ISDIR (st.st_mode);else ;/* punt */
With Carlo's patch the following tests fail:
$ sh t4053-diff-no-index.sh
[...]
not ok 35 - diff --no-index with pathspec nested pathspec
#
# test_expect_code 1 git diff --name-status --no-index c d 1/2 >actual &&
# cat >expect <<-EOF &&
# D c/1/2/a
# D c/1/2/b
# EOF
# test_cmp expect actual
#
not ok 36 - diff --no-index with pathspec glob
#
# test_expect_code 1 git diff --name-status --no-index c d ":(glob)**/a" >actual &&
# cat >expect <<-EOF &&
# D c/1/2/a
# EOF
# test_cmp expect actual
#
ok 37 - diff --no-index with pathspec glob and exclude
# failed 2 among 37 test(s)But with your patch (+ the minor corrections) the tests pass as expected:
$ sh t4053-diff-no-index.sh
[...]
ok 35 - diff --no-index with pathspec nested pathspec
ok 36 - diff --no-index with pathspec glob
ok 37 - diff --no-index with pathspec glob and exclude
# passed all 37 test(s)Therefore, I think your fix is good to go with a commit message and my changes. Feel free to add me to Reviewed-by/Tested-By.
Collin