Re: Regression with large XML data input

Michael Paquier <michael(at)paquier(dot)xyz> writes:
> On Thu, Dec 25, 2025 at 12:50:38PM -0500, Tom Lane wrote:
>> That would be my preferred answer, but after a quick census of the
>> buildfarm I'm not sure we can get away with it. The animals that
>> are showing these warnings are all running RHEL 10 or clones of it,
>> which is going to be in-support till 2035 or so. Maybe Red Hat
>> will upgrade to a newer libxml2 version than whatever they chose
>> for RHEL 10.0, but I would not count on it.

> Hmm. Having to deal with these warnings until 2035 kind of sucks. Do
> you think that it be worth forcing the use of
> -Wno-deprecated-declarations only for versions older than 2.13.3 in
> the Makefiles where we could see these warnings?

It seems pretty painful to identify the libxml2 version during
configure, do you have an idea how to do that? (And how about
on the meson side?)

> As a whole, this kind of enforcement gives me mixed feelings, because
> it would also mean that we would miss what looks like legit
> deprecation warnings that we had better look at and study, even for
> LLVM. At least not enforcing any policy lets us know about these
> issues.

Right. I feel very itchy about adding -Wno-deprecated-declarations
with no plan for getting rid of it again. Seeing that libxml2 seems
to be a moving target again, it seems like that would eventually bite
us on the rear.

[ thinks... ] Actually, I believe we do have precedents in configure
for checking to see if a construct produces a warning, so we could
check to see if a reference to xmlKeepBlanksDefault draws one, and
only add -Wno-deprecated-declarations if it does. But I'm not sure
how to do that in meson.

regards, tom lane