Re: [Nmh-workers] Proposed solution to Debian Bug#143485

[Harald Geyer]

>           execlp (mhlproc, mhl, "-form", filter, "-noclear", NULL);
> -         fprintf (stderr, "unable to exec ");
> -         perror (mhlproc);
> +         write(2, "unable to exec ", 15);
> +         write(2, mhlproc, strlen(mhlproc));
> +         write(2, strerror(errno), strlen(strerror(errno)));
> 
> The code you changed there is obviously broken (apart from using printf after
> a vfork), but your replacement isn't a lot better - the "strerror(errno)"
> calls (which you really should not do twice ... use a variable and save the
> result!) the way you have it is getting the errno set from the write(2,
> mhlproc, strlen(mhlproc)) sys call, where I suspect that you wanted the
> errno from the execlp (which must have failed or the code wouldn't be being
> executed).

Hm, I think if write(2, ...) fails, then we are out of luck anyway. We
can't give an error message about being unable to give an error message.
I thought that a successful call of write won't manipulate errno, but after
rereading the documentation of errno it seems that one can't rely on this.
On the other hand it says, that one can detect an error by explictely
setting errno=0 before the call and testing errno afterwards. It also
says that errno is never set to zero by the library. So I'm now a bit
confused about how write is allowed to change errno on success...

> write this as
>       char *errstr;
>       [...]
>       execlp(...);
>       errstr = strerror(errno);
>       write(2,....);
>       write(2,....);
> /* maybe: write(2, ": ", 2); */
>       write(2, errstr, strlen(errstr));
>       write(2, "\n", 1);

I agree that this is a better solution. If anybody is going to commit my
patch, he hopefully will incorporate this.

Grüße,
Harald