svn commit: r326685 - in head: devel/mercator math/wfmath

[Alexey Dokuchaev]

On Sun, Sep 08, 2013 at 11:52:53AM +0200, Oliver Lehmann wrote:
> Alexey Dokuchaev <danfe at freebsd.org> wrote:
> >unclear is 1) what kind of error this commit is supposed to fix;
> 
> USE_GCC was introduces because: It requires a gcc compiler to compile.
> Next time I'll think more about the message to make this more clear.

It was quite obvious from the diff that GCC was pulled to unbreak the build
with Clang; my concern was that is was not obvious what exactly was wrong,
and thus it was not clear if USE_GCC was used wisely (instead of a proper
fix).  Excerpt from the build log (just few lines) would be enough if you
do not have/cannot provide sufficient analysis of the problem at the time.

> >2) why is it hard to create a proper patch instead;
> 
> Why? I can't tell, I'm sorry - I just was not able to fix the clang
> compilation problems otherwise.

Again, it's hard to understand that without the documenting why USE_GCC was
used in the first place.  Second, given the suggestion of simply removing
misused CONFIGURE_ARGS (and provided that it does indeed solve the issue),
I find it a bit hard to believe you could not have come up with it yourself
Oliver. ;-)

> >3) is it really necessary to set USE_GCC to "yes", not "any"?
> 
> This is what the header of bsd.gcc.mk suggests. If "any" is prefered over
> "yes", it should be stated somewhere in the top of bsd.gcc.mk - maybe in
> the Examples section which is where I always look first.

"yes" and "any" do different things, it's not the matter of style.  "yes"
means to use modern GCC compiler, even on systems that still use GCC as
a system one by default (e.g. 8.x, 9.x).  "any" means that literally any
GCC version would suffice, which means that on our stable branches, I do
not have to install some GCC 4.6+ since I already gave GCC 4.2 in the base.

./danfe