RE: [Fwd: Debugging execve problems]

[Svante Signell]

On Wed, 2012-01-04 at 21:44 +0000, Svante R Signell wrote:
> Samuel,
> 
> I'm replying with M$ webmail, therefore the top posting, sorry for that. My 
> main mailer is currently broken.
> 
> Regarding the execve stuff, where can I find the bug and the patch to these 
> problems?

Found the email discussion on the bug-hurd mailing list from May-June
2010.

> Svante Signell, le Wed 04 Jan 2012 16:44:20 +0100, a écrit :
> > #!/bin/sh
> > IXBIN=/home/srs/DEBs/icon/debugging_icon/iconx
> > IXLCL=`echo $0 | sed 's=[^/]*$=iconx='`
> 
> Ah, the executed thing is a shell script, not a binary. Ok, it's really
> the same issue as Emilio was fixing some time ago, and reduces to this:
> 
> test.c:
> int main(int argc, char *argv[]) {
>         char *t[] = {"test.sh", NULL};
>         execv("test.sh", t);
>         perror("execve");
>         return 0;
> }
> 
> test.sh:
> #!/bin/sh
> echo I'm $0
> 
> test.sh gets /dev/fd/3, which is not helpful for the IXLCL computation
> above.

Some questions related to the above:

1) I understand that if Emilos patches introducing a new RPC would be ABI and 
API changes for
all architectures there would be a problem, are they really? 
2) How come execve calls a shell script works with . in PATH?
3) How come other architectures don't have this problem, and only Hurd
has?