Re: Defend against -ffast-math in meson builds

Hi,

On Wed, Mar 11, 2026 at 01:48:31PM -0400, Andres Freund wrote:
> Hi,
>
> On 2026-03-11 12:45:55 -0400, Tom Lane wrote:
> > So no, I don't wanna support this. But maybe we should move the
> > code-level tests out of the datetime files and into utils/float.h
> > or some such place.
>
> I think it's probably better to have it in a .c file (maybe float.c), I could
> kinda imagine some extension intentionally enabling -ffast-math, because it
> does something numerically intensive where the incorrectness doesn't matter.

I think that you have a good point about the extension. That said a .h file
could also prevent the extension to make use of -ffast-math "accidentally".

And if they really want it, I think that they could compile the extension with
"-ffast-math -U__FAST_MATH__" to make it work. I just tested it and that works
but now that I write it that looks like an odd combination, so a .c file looks
better.

What do you think?

Regards,

--
Bertrand Drouvot
PostgreSQL Contributors Team
RDS Open Source Databases
Amazon Web Services: https://aws.amazon.com