RE: Fixes to config.log handling

[Tim Van Holder]

> The first two hunks make sure we only write to config.log through
> the fd; otherwise the file got corrupted (on DOS at least), probably
> because config.log was opened in append mode.  I'm also opening it
> normally, because I _think_ the idea is for config.log is for a single
> run of configure + any and all runs of the config.status produced by
> that configure.
> The last hunk closed the fd for config.log (by redirecting to /dev/null)
> before running config.status and reopens it afterwards.  This is needed
> on DOS; otherwise, config.status will fail to write to config.log (either
> because that fd is already in use, or because the output file is still
> open).
> If it would cause trouble on other platforms, the last hunk can be
> omitted for now (it can always be added to the devel branch), but I think
> the first two are generally useful (and should probably make it 
> into 2.50).

The last hunk was broken; it caused the exit status of config.status to be
lost, which broke several tests.  The attached diff fixes this.

2001-04-14  Tim Van Holder  <address@hidden>

        * acgeneral.m4: Always use the file descriptor to write to
        config.log to avoid corruption.  Close the descriptor before
        running config.status so config.log is properly created on
        MS-DOS.

config-log-issues2.diff
Description: Binary data