[gentoo-dev] Re: [RFC] New policy: LDFLAGS should be respected

[Nikos Chantziaras]

Ciaran McCreesh wrote:
> On Sun, 27 Jul 2008 00:00:55 +0200
> Carsten Lohrke <carlo@g.o> wrote:
>> Afaik it has always been the way that *sane* LDFLAGS are to be
>> respected, but exceptions exist of course and it's up to the
>> maintainer to mangle or clear your LDFLAGS, if deemed necessary. I'd
>> like to know, why Mark asked to bring this question up here.
>> Shouldn't this be common sense!?
>
> The way it is currently: Packages ignoring CFLAGS without a *very* good
> reason (and 'upstream thinks they know better' is rarely a very good
> reason, especially when upstream supposedly knowing better leads to v7
> builds on v9 systems) need to be fixed. Packages ignoring LDFLAGS can
> be fixed if the maintainer feels like it, but there's no requirement to
> do so and filing bugs about it is frowned upon.
>
> Until recently, LDFLAGS have been put in the "anyone using these is a
> ricer" category. Unfortunately, the misguided promotion of 'as-needed'
> despite its massive design flaws has lead people to think that setting
> LDFLAGS is in some way useful or cool. I expect next someone will try
> to find a way to force 'ASFLAGS' onto everyone...

This is totally irrelevant though.  If I have --as-needed in my LDFLAGS 
(I do) I still consider it a bug if a package does not honor it.  So 
what I'm doing is fixing the ebuild (*if* the ebuild does not mention a 
reason of not honoring LDFLAGS of course) and submit it in bugzilla.  I 
don't know if the maintainers are getting annoying by this.  They 
shouldn't.  If some LDFLAGS turn out to break a package in some way 
doesn't mean that it's OK for the package to ignore LDFLAGS altogether. 
  If I have CFLAGS="-O999999 -fsuper-mega-fast-math 
-enable-leet-broken-experimental-optimize" doesn't mean the package 
should ignore CFLAGS :P

(As for --as-needed, it's the same as -O3 in CFLAGS; if a package turns 
out to break, an ebuild *could* explicitly filter out -O3, but that's 
not really a priority.  In the end, if the user chooses ricer-flags and 
breaks his system, he can blame himself.  If he explicitly wants to 
shoot himself in the foot, the ebuild should allow him to do so.  Any 
effort spent to protect the ricers from themselves is *wasted* effort 
better spent somewhere else.

Not that I have ever seen a package that breaks with --as-needed though. 
  Of course that's just me.)