Andres Freund <andres@2ndquadrant.com> writes:
> On 2015-01-13 22:19:30 -0500, Tom Lane wrote:
>> The reason I got interested in this is that I attempted to pass in
>> "CFLAGS=-Wno-format" to configure, to suppress format warnings on
>> buildfarm member gaur (whose gcc is too old to recognize z modifiers).
>> That doesn't work because -Wall turns the warnings right back on again.
>> If the user-supplied CFLAGS were inserted after -Wall then it would work.
> In the line of
> http://archives.postgresql.org/message-id/54B58BA3.8040302%40ohmu.fi in
> wonder if the better fix isn't to define pg_format_attribute(...) and
> define it empty that if the compiler doesn't support what we want.
Well, that would only fix my problem if we added a configure-time test
for whether gcc recognizes "z", which frankly seems like a waste of
cycles. I've probably got the last one left in captivity that doesn't.
Not that I have any great objection to improving the attribute-slinging
logic. It just doesn't seem like a good fix for this particular issue.
>> A slightly more complicated change could be applied to make sure that
>> *all* of the CFLAGS forcibly inserted by configure appear before any
>> externally-sourced CFLAGS, allowing any of them to be overridden from the
>> environment variable. I'm not sure if it's worth the trouble to do that,
>> but if there's interest I could make it happen.
> I think it'd be good idea, but unless you're enthusiastic I guess there
> are more important things.
Nah, I'm fine with doing it, it's just a couple more lines of code.
I feared people might think it wasn't worth adding extra complexity for,
but as long as someone else likes the idea I'll go do it.
regards, tom lane