On 08/01/2014 22:28, Chandler Carruth wrote:> On Wed, Jan 8, 2014 at 6:38 AM, Alp Toker <alp at nuanti.com
> <mailto:alp at nuanti.com>> wrote:
>
> (The examples should really be enabled by default, not just to aid
> testing but
> also to prevent bitrot in some key user-facing code. I'll propose
that
> shortly.)
>
>
> Just as an FYI as I remember the discussion but can't find it in email
> any more: the example building was turned off some time ago to speed
> up incremental development. It was kept enabled by bots (in theory,
> not sure how this worked out in practice) while being disabled by default.
>
> I've never completely understood the build performance concern, but ::
> shrug ::.
Good point. I remember people complaining about that too.
Would it be possible to have the examples built the same time as the
UnitTests in !EXAMPLES configurations so the 'all' target remains
'fast'
for the complainers?
>
> I'd also be perfectly happy to make a test-only clang plugin that
> isn't an example. We already have test-specific tools that seem quite
> similar. Your call.
Thanks! That'll make sense if begin to test more features that clutter
the sample code. So my arbitrary watershed for forking is "when the
example stops serving as clear and useful sample code." Right now we're
in the sweet spot though, and I'm enjoying having examples that are
tested given how broken they've historically been.
If you have a minute, my suggestion would rather be the LLVM side. It
should be a matter of proposing the switch back to ".so" from
".dylib"
and enabling MODULE in the CMake files to fix PR14903 .
A straight revert of r122395 (which was bogus) should do the trick to
get LLVM back on side.
Alp.
--
http://www.nuanti.com
the browser experts