Re: [PATCH] config.mak.dev: enable -Wpedantic in clang

From: Junio C Hamano <gitster@pobox.com>
Date: Nov 29, 2018, 05:13 UTC
Message-ID: <xmqqva4gpits.fsf@gitster-ct.c.googlers.com>
In-Reply-To: <CAPig+cQXkNo_Sztf3y5ciZ3jjj=XrJ+ESCxEsU_RKxb_0x+STw@mail.gmail.com>

Eric Sunshine <sunshine@sunshineco.com> writes:

Show 8 quoted lines

> Playing Devi's Advocate, what if Apple's clang "8" was, in reality,
> real-world clang 3? Then this condition would incorrectly enable the
> compiler option on Apple for a (real) clang version below 4. For this
> reason, it seems we shouldn't be trusting only the clang version
> number when dealing with Apple.
>
> (I suspect that it won't make a difference in actual practice, so it
> may be reasonable to punt on this issue until/if someone complains.)

Why do we care which true version of clang is being used here in the first place? Is it because some version of clang (take -Wpedantic but misbehave | barf when -Wpedantic is given) while some others are OK?

If the only problem is that some version of clang barf when the option is given, perhaps we can do a trial-compile of helloworld.c or something, or is that something we are trying to avoid in the first place?

It appears to me that ./detect-compiler tool (by the way, perhaps we should start moving these build-helper-tools sitting at the top level down to ./build-helpers/ subdirectory or something so that we can focus on the source code when running "ls" at the top level of the hierarchy) can become more intelligent to help things like this.