Hi Ben,
Thanks for passing the patch upstream. Sorry for finding issues with it
now.
* Ben Pfaff wrote on Tue, Aug 12, 2008 at 07:14:09AM CEST:
> diff --git a/bin/autom4te.in b/bin/autom4te.in
> old mode 100644
> new mode 100755
Why this mode change?
> index 685df41..1ea86da
> --- a/bin/autom4te.in
> +++ b/bin/autom4te.in
> @@ -550,21 +550,36 @@ sub handle_output ($$)
> else
> {
> - $out->open($output, O_CREAT | O_WRONLY | O_TRUNC, oct ($mode));
> + $out->open ("$output.tmp", O_CREAT | O_WRONLY | O_TRUNC, oct ($mode));
Hmm. This may still leave us competing with a concurrent autoconf
running. I'm not saying that this is something we need (there are
more places that need fixing in order to guard against this), but
shouldn't a temporary file have a temp name (or at least using
a PID or so)?
We use mktmpdir earlier, but possibly on a different mount, which
will break the atomicity if we create the file there.
> @@ -600,6 +615,12 @@ sub handle_output ($$)
>
> $out->close();
>
> + if ($atomic_replace && !rename ("$output.tmp", "$output"))
> + {
> + move ("${output}.tmp", "$output")
Please use a variable for "${output}.tmp".
> + or fatal "cannot rename ${output}.tmp as $output: $!";
How about still trying to remove the temp file?
A test case to expose the issue fixed by this patch would be nice
(but I'm not sure how to do it easily).
Thanks,
Ralf
