Обсуждение: Warning-suppression fixes we ought to back-patch

I looked through the buildfarm's compiler warnings on back branches,
which is something I rarely do, and noted a bunch of warnings that
we are not seeing on master.  That's because of recent commits that
aimed to silence those warning types.  Most of the noise could be
suppressed by back-patching these two changes:

0909380e4    Allow PG_PRINTF_ATTRIBUTE to be different in C and C++ code.

This only matters for builds that are combining gcc with clang++ or
vice versa, but we have a dozen or so BF animals that are like that,
and each one is spewing a hundred "unrecognized format function type"
warnings.

8f1791c61    Fix some cases of indirectly casting away const.

While we've done a lot of const-cleanliness work recently, this patch
should be enough to silence most of the cast-away-const warnings I'm
seeing in the BF back branches.  Only "midge" is showing them at the
moment, but more people will be seeing them as gcc 16 gets more
widespread.  (I'd probably skip the ecpg bits though, as they are
a bit more convoluted than the rest, so the risk/benefit ratio
seems poor.)

Both of these patches were applied only to master, but with the
expectation that we'd back-patch once they'd baked awhile.  It's
been two-plus months which seems like enough.  So barring objections,
I will back-patch them once the current release freeze is over.

            regards, tom lane

Hi,

On Tue, Feb 24, 2026 at 11:24:45AM -0500, Tom Lane wrote:
> 
> 8f1791c61    Fix some cases of indirectly casting away const.

Worth to also back-patch 9f7565c6c2d (as 8f1791c61 follow up)?

Regards,

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

On 24.02.26 17:24, Tom Lane wrote:
> I looked through the buildfarm's compiler warnings on back branches,
> which is something I rarely do, and noted a bunch of warnings that
> we are not seeing on master.  That's because of recent commits that
> aimed to silence those warning types.  Most of the noise could be
> suppressed by back-patching these two changes:
> 
> 0909380e4    Allow PG_PRINTF_ATTRIBUTE to be different in C and C++ code.
> 
> This only matters for builds that are combining gcc with clang++ or
> vice versa, but we have a dozen or so BF animals that are like that,
> and each one is spewing a hundred "unrecognized format function type"
> warnings.
> 
> 8f1791c61    Fix some cases of indirectly casting away const.

No problem backpatching these.

> While we've done a lot of const-cleanliness work recently, this patch
> should be enough to silence most of the cast-away-const warnings I'm
> seeing in the BF back branches.  Only "midge" is showing them at the
> moment, but more people will be seeing them as gcc 16 gets more
> widespread.  (I'd probably skip the ecpg bits though, as they are
> a bit more convoluted than the rest, so the risk/benefit ratio
> seems poor.)

These warnings came in from the new glibc version 2.43, not from the 
compiler.

Bertrand Drouvot <bertranddrouvot.pg@gmail.com> writes:
> On Tue, Feb 24, 2026 at 11:24:45AM -0500, Tom Lane wrote:
>> 8f1791c61    Fix some cases of indirectly casting away const.

> Worth to also back-patch 9f7565c6c2d (as 8f1791c61 follow up)?

Good idea.

            regards, tom lane

Peter Eisentraut <peter@eisentraut.org> writes:
> On 24.02.26 17:24, Tom Lane wrote:
>> While we've done a lot of const-cleanliness work recently, this patch
>> should be enough to silence most of the cast-away-const warnings I'm
>> seeing in the BF back branches.  Only "midge" is showing them at the
>> moment, but more people will be seeing them as gcc 16 gets more
>> widespread.

> These warnings came in from the new glibc version 2.43, not from the 
> compiler.

Ah.  Although I suppose you also need a sufficiently late-model
compiler, to understand the annotations on strchr() and friends.
Anyway, thanks for the correction.

            regards, tom lane