Re: Mixing CC and a different CLANG seems like a bad idea

Andres Freund <andres(at)anarazel(dot)de> writes:
> On 2021-11-18 13:39:04 -0500, Tom Lane wrote:
>> More, I think (though this ought to be documented in a comment) that
>> the policy is to not bother turning on extra -W options in the bitcode
>> switches, on the grounds that warning once in the main build is enough.
>> I follow that idea --- but what we missed is that we still need to
>> turn *off* the warnings we're actively disabling. I shall go do that,
>> if no objections.

> Thanks for doing that, that does sounds like a good way, at least for now.

Cool, thanks for confirming.

For the archives' sake: I thought originally that this was triggered
by having CC different from CLANG, and even wrote that in the commit
message; but I was mistaken. I was misled by the fact that sidewinder
is the only animal still reporting the compound-token-split-by-macro
warnings, and jumped to the conclusion that its unusual configuration
was the cause. But actually that't not it, because the flags we
feed to CLANG are *not* dependent on what CC will take. I now think
the actual uniqueness is that sidewinder is the only animal that is
using clang >= 12 and has --with-llvm enabled.

>> (BTW, does meson have any comparable optimization?
>> If it doesn't, I bet that is going to be a problem.)

> Yes - imo in a nicer, more reliable way. It caches most test results, but with
> the complete input, including the commandline (i.e. compiler flags) as the
> cache key. So no more errors about compile flags changing...

Nice!

regards, tom lane