Re: [9.0beta5/cvs head] build failure due to unchecked results

On Fri, Apr 30, 2010 at 9:14 AM, Peter Eisentraut <peter_e(at)gmx(dot)net> wrote:
> On fre, 2010-04-30 at 14:43 +0200, Andres Freund wrote:
>> On Friday 30 April 2010 13:56:11 Peter Eisentraut wrote:
>> >
>> > You probably mean alpha5, unless you come from the future. ;-)  That was
>> > actually a mistake in the packaging, which is why there is a -revised
>> > tarball available.
>> Isnt it a good idea to set this in the pre-release versions? It might catch
>> potential problems and people trying to compile unreleased software from
>> source should be able to change it/report a bug if necessary.
>
> In practice it might prevent more people from building and thus testing
> PostgreSQL than it would catch actually interesting problems.  You also
> can't assume that the same warning-freeness applies to all platforms.

Yeah, that's the real problem - one doesn't get the same warnings
everywhere. Still, I think we should consider applying the portion of
the proposed patch that avoid relying on the contents of the fgets()
buffer after fgets() returns NULL, because at least on MacOS the man
page for gets/fgets says:

If an error occurs, they return NULL and the buffer contents are indeterminate.

Indeterminate is bad. The actual chance of a crash seems low, and I
notice that the Linux man page has no such caveat, but still.

...Robert