On Sun, Mar 22, 2020 at 8:31 AM Torsten Bögershausen <tboegi@xxxxxx> wrote: > > On Sat, Mar 21, 2020 at 12:42:50PM +0100, András Kucsma wrote: > > Hi All, > > > > I believe to have found an issue regarding properly executing the > > GIT_ASKPASS binary. I'm using Windows Server 2019, with git 2.21.0 > > installed using cygwin. > > > > ## To reproduce: > > > > Assume you have the askpass binary at C:\askpass.bat. In CMD the > > following commands reproduce the issue: > > > > C:\> set GIT_ASKPASS=C:\askpass.bat > > C:\> git clone https://<private_repository>.git > > Cloning into '<private_repository>'... > > error: cannot run C:\askpass.bat: No such file or directory > > [... proceeds to interactively ask for username and password ...] > > > > On the other hand, if we change the GIT_ASKPASS environment variable > > slightly, so that there is a forward slash (/) instead of a backslash > > (\), things work as expected: > > > > C:\> set GIT_ASKPASS=C:/askpass.bat > > C:\> git clone https://<private_repository>.git > > Cloning into '<private_repository>'... > > [... success ...] > > > > ## Some context: > > > > The source of the problem, is that if git doesn't find a forward slash > > anywhere in the path, it assumes it is not a real path and has to look > > for the binary using the PATH environment variable. See in > > prepare_cmd(): > > https://github.com/git/git/blob/98cedd0233e/run-command.c#L429-L439 > > > > You can see that the "cannot run" error message is printed here, just > > after prepare_cmd() returned -1: > > https://github.com/git/git/blob/98cedd0233e/run-command.c#L749-L753 > > > > I believe this was introduced in late 2018 around git v2.19.2, > > although I did not actually bisect the issue: > > https://github.com/git/git/commit/321fd823897#diff-7577a5178f8cdc0f719e580577889f04R401-R415 > > > > > > I hope I'm sharing this bug at the right forum. Please direct me to > > the proper place if not. > > Yes, you came to the rigth place. > Thanks for the report and the detailed analysis. > > A quick fix, and a begin of a patch, could be to use > has_dos_drive_prefix() which will look for C: and will therefore even work > with C:\ > > /* > * If there are no '/' characters in the command then perform a path > * lookup and use the resolved path as the command to exec. If there > * are '/' characters, we have exec attempt to invoke the command > * directly. > */ > if ((!strchr(out->argv[1], '/')) || > (has_dos_drive_prefix(out->argv[1]))) { > char *program = locate_in_PATH(out->argv[1]); > [] > > If you want to play around with the code a little bit, and send us a "git diff", > we can convert that into a patch. > > Wellcome to the Git community. > > > > > Thank you, > > Andras Thanks Torsten! I believe it is not enough to test only for the drive specifier, as GIT_ASKPASS has to work with relative paths as well: C:\SomeDirectory> set GIT_ASKPASS=.\SomeOtherDirectory\askpass.bat C:\SomeDirectory> git clone https://<some_private_repository>.git My proposal patch is to take advantage of find_last_dir_sep function's OS specific directory separator knowledge. I posted the diff below, which is also available on github here: https://github.com/git/git/compare/maint...r0mai:fix-prepare_cmd-windows-maint diff --git a/run-command.c b/run-command.c index f5e1149f9b..9fcc12ebf9 100644 --- a/run-command.c +++ b/run-command.c @@ -421,12 +421,12 @@ static int prepare_cmd(struct argv_array *out, const struct child_process *cmd) } /* - * If there are no '/' characters in the command then perform a path - * lookup and use the resolved path as the command to exec. If there - * are '/' characters, we have exec attempt to invoke the command - * directly. + * If there are no dir separator characters in the command then perform + * a path lookup and use the resolved path as the command to exec. If + * there are dir separator characters, we have exec attempt to invoke + * the command directly. */ - if (!strchr(out->argv[1], '/')) { + if (find_last_dir_sep(out->argv[1]) == NULL) { char *program = locate_in_PATH(out->argv[1]); if (program) { free((char *)out->argv[1]);
