mingw: take argv[0] from command line parameters

[kgybels]

Instead of basing argv[0] on the executable path, use argv[0] as
provided on the command line.

Setting argv[0] to git- is currently the only way to prevent
git from falling back to calling a helper executable. This change will
allow the git-wrapper executable, that Git for Windows uses, to signal
to git to handle the command as a builtin.

Related to #1496

[kgybels]

@dscho Initially I wanted to send this to the git mailing list, however, I noticed that mingw_startup is in git/git but msc_startup is not.

[mfriedrich74]

GetModuleFileNameW is the function that I would recommend to use. That returns the path for dlls or the current executable as absolute path.

…

On Thu, Mar 15, 2018, 8:54 PM kgybels ***@***.***> wrote: ***@***.**** commented on this pull request. ------------------------------ In compat/mingw.c <#1561 (comment)>: > @@ -3396,8 +3396,7 @@ int msc_startup(int argc, wchar_t **w_argv, wchar_t **w_env) */ ALLOC_ARRAY(my_utf8_argv, argc + 1); ALLOC_ARRAY(save, argc + 1); - save[0] = my_utf8_argv[0] = wcstoutfdup_startup(buffer, _wpgmptr, maxlen); Just now, before I saw this comment, I ran into a crash because of it at exec_cmd.c#L20 <https://github.com/git/git/blob/v2.16.2/exec_cmd.c#L20>, when I did: C:\git\bug\test>C:\PortableGit\mingw32\libexec\git-core\git-status.exe I see that it needs git_extract_argv0_path() <https://github.com/git/git/blob/v2.16.2/exec_cmd.c#L31> to set up argv0_path, which was still NULL causing the crash. It seems wrong to me to take the executable path and put it in argv[0], to later use argv[0] to calculate the system prefix. If we want to know our system prefix relative to our executable, then use the executable path directly instead of using argv[0]. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#1561 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AHJ-yKMmG9ZBFWz9r8b09o12poXS8_Jeks5tew1bgaJpZM4Ss_rd> .

[dscho]

GetModuleFileNameW is the function that I would recommend to use. That returns the path for dlls or the current executable as absolute path.

Thanks for the suggestion! We already use _wpgmptr which serves the same purpose ;-)

[danjacques]

My apologies for delay - I paused that patch set pending a Perl Makefile refactor, which landed. I have a rebased version almost ready to go, but haven't been prioritizing it. Since there is some interest and I'm blocking people, I'll try and find some time to push another set out soon. Sound good?

…

On Fri, Mar 16, 2018, 8:51 AM kgybels ***@***.***> wrote: ***@***.**** commented on this pull request. ------------------------------ In compat/mingw.c <#1561 (comment)>: > @@ -3396,8 +3396,7 @@ int msc_startup(int argc, wchar_t **w_argv, wchar_t **w_env) */ ALLOC_ARRAY(my_utf8_argv, argc + 1); ALLOC_ARRAY(save, argc + 1); - save[0] = my_utf8_argv[0] = wcstoutfdup_startup(buffer, _wpgmptr, maxlen); Any idea what effort remains for getting exec_cmd: RUNTIME_PREFIX on some POSIX systems stable enough? I was refactoring in a similar direction ( ***@***.*** <kgybels@dca5afd>), however, @danjacques <https://github.com/danjacques>'s work is much more evolved. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#1561 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AJgOMBchIyyKKCiQrymdxDJRkoa6O8FKks5te7VcgaJpZM4Ss_rd> .

[dscho]

Sound good?

@danjacques very!

[kgybels]

@danjacques Sounds great!

@dscho I will wait for the rebased version, then add my work on top.

[kgybels]

@dscho and @danjacques already did all the work (6e007ce), so I'm closing this pull request.