Support code coverage.

Support code coverage.

COMMIT=false

[Camillo Bruni, 2015-08-17 05:24:16]

The CQ bit was checked by cbruni@chromium.org

[Camillo Bruni, 2015-08-17 05:24:17]

The CQ bit was unchecked by cbruni@chromium.org

[Camillo Bruni, 2015-08-17 05:24:36]

The CQ bit was unchecked by cbruni@chromium.org

[Michael Achenbach, 2015-08-17 07:37:08]

If you want to land a CL like this, please go ahead. This was not ready yet. I'd
wanna add the following things:
- Make coverage=1 imply clang=0
- Properly unset any -O1/2/3 cflags - I think right now we end up with two flags
set, e.g. "-O3 -O0", which is a bit messy
- Figure out if we need to unset any other optimization options
- Make sure that the host_clang option is not in the way

If nobody else picks it up, I'll work on it when I'm back from my leave...

[Michael Achenbach, 2015-08-17 07:45:28]

On 2015/08/17 07:37:08, Michael Achenbach (OOO) wrote:
> If you want to land a CL like this, please go ahead. This was not ready yet.
I'd
> wanna add the following things:
> - Make coverage=1 imply clang=0
> - Properly unset any -O1/2/3 cflags - I think right now we end up with two
flags
> set, e.g. "-O3 -O0", which is a bit messy
> - Figure out if we need to unset any other optimization options
> - Make sure that the host_clang option is not in the way
> 
> If nobody else picks it up, I'll work on it when I'm back from my leave...

Probably coverage=1 should also imply host_clang=0. Otherwise the host d8 would
be compiled with clang and we'd have no code coverage for the snapshot
generation...

[Michael Achenbach, 2015-08-17 07:45:29]

On 2015/08/17 07:37:08, Michael Achenbach (OOO) wrote:
> If you want to land a CL like this, please go ahead. This was not ready yet.
I'd
> wanna add the following things:
> - Make coverage=1 imply clang=0
> - Properly unset any -O1/2/3 cflags - I think right now we end up with two
flags
> set, e.g. "-O3 -O0", which is a bit messy
> - Figure out if we need to unset any other optimization options
> - Make sure that the host_clang option is not in the way
> 
> If nobody else picks it up, I'll work on it when I'm back from my leave...

Probably coverage=1 should also imply host_clang=0. Otherwise the host d8 would
be compiled with clang and we'd have no code coverage for the snapshot
generation...

[Camillo Bruni, 2015-08-17 08:04:46]

On 2015/08/17 07:45:29, Michael Achenbach (OOO) wrote:
> On 2015/08/17 07:37:08, Michael Achenbach (OOO) wrote:
> > If you want to land a CL like this, please go ahead. This was not ready yet.
> I'd
> > wanna add the following things:
> > - Make coverage=1 imply clang=0
> > - Properly unset any -O1/2/3 cflags - I think right now we end up with two
> flags
> > set, e.g. "-O3 -O0", which is a bit messy
> > - Figure out if we need to unset any other optimization options
> > - Make sure that the host_clang option is not in the way
> > 
> > If nobody else picks it up, I'll work on it when I'm back from my leave...
> 
> Probably coverage=1 should also imply host_clang=0. Otherwise the host d8
would
> be compiled with clang and we'd have no code coverage for the snapshot
> generation...

I'm aware that this is not ready yet in a clean way :) (and I know too little
about the build system for now). Maybe we can land this for now so people can
try it locally, and then later get the real change in probably with a separate
.coverage target etc. for the bots.