Re: [gentoo-dev] [PATCH 2/2] eclass/toolchain-funcs: add clang version functions

[Michał Górny]

On Mon, 4 Jul 2016 19:16:55 -0500
Austin English <wizardedit@g.o> wrote:

> On 07/01/2016 03:02 PM, Michał Górny wrote:
> > On Mon, 27 Jun 2016 17:04:41 -0500
> > Austin English <wizardedit@g.o> wrote:
> >
> >> From ec0be3d1a808ea0c5bdd081a4bb935f86bf78d44 Mon Sep 17 00:00:00 2001
> >> From: Austin English <wizardedit@g.o>
> >> Date: Mon, 27 Jun 2016 16:58:07 -0500
> >> Subject: [PATCH 2/2] eclass/toolchain-funcs: add clang version functions
> >
> > Why do you even ask for review if you can't wait till the weekend
> > before committing?
>
> There's no set time limit that I could find anywhere. I checked with my
> mentor first, who said a few days was enough (as there were no comments
> by anyone).
>
> > I've already told you twice that I'm opposed to adding version querying
> > functions for clang, especially for the use case you proposed. You
> > should check for features, and not keep a huge list of supported gcc,
> > clang, pathcc, icc... versions. With every random compiler pretending
> > to be a random version of every other compiler.
>
> I did not use in the original intended case, however I think that these
> helpers are still useful. Why do we allow gcc version checks then?

Because the ebuilds are full of that nonsense, and developers using
them don't help you remove it.

If we committed every single thing someone thought someone else might
find useful one day, the eclass would be huge hogs full of useless code
that is only used because someone noticed the function and suddenly
came to conclusion it's a good idea to use it because someone added it.

Now let's add 4 additional functions for every single compiler that
someone may decide to support one day. I'm even strongly opposed to
adding functions that are used only theoretically.

> There's no deprecated comment for these functions. By that logic, you
> should not have added tc-is-{gcc,clang}, 2 weeks ago in
> 6984a5b149a215dd96a9759d3d1f251354faf38f. You should be testing for
> compiler features directly, not keep a list of what compilers are
> supported in the code...

Yes, I can revert this if you insist. And don't mind that all those
horribly broken ebuilds checking gcc versions will suddenly fail with
clang.

> The point of eclasses is to share code and make ebuild maintenance
> easier. I don't see how allowing version checks for GCC only, when clang
> is supported by a lot of ebuilds, makes that easier. Again, if you're so
> opposed to compiler specific checks, then please remove them from
> ebuilds using them and from toolchains-funcs.eclass (or at least mark
> them as deprecated).

How does ebuild maintenance become easier when you have to test a dozen
of versions of different compilers to figure out which one is
supported?

-- 
Best regards,
Michał Górny
<http://dev.gentoo.org/~mgorny/>