r232439 - GCOV: Expose the -coverage-exit-block-before-body flag in clang -cc1

[Nick Lewycky]

David Blaikie wrote:
>
>
> On Mon, Mar 16, 2015 at 6:18 PM, Nick Lewycky <nlewycky at google.com
> <mailto:nlewycky at google.com>> wrote:
>
>     On 16 March 2015 at 18:06, Justin Bogner <mail at justinbogner.com
>     <mailto:mail at justinbogner.com>> wrote:
>
>         David Blaikie <dblaikie at gmail.com <mailto:dblaikie at gmail.com>>
>         writes:
>         >  On Mar 16, 2015 4:57 PM, "Justin Bogner"
>         <mail at justinbogner.com <mailto:mail at justinbogner.com>> wrote:
>         > >
>         > > Author: bogner
>         > > Date: Mon Mar 16 18:52:21 2015
>         > > New Revision: 232439
>         > >
>         > > URL: http://llvm.org/viewvc/llvm-project?rev=232439&view=rev
>         <http://llvm.org/viewvc/llvm-project?rev=232439&view=rev>
>         > > Log:
>         > > GCOV: Expose the -coverage-exit-block-before-body flag in
>         clang -cc1
>         > >
>         > > This exposes the optional exit block placement logic from
>         r232438 as a
>         > > clang -cc1 option. There is a test on the llvm side, but
>         there isn't
>         > > really a way to inspect the gcov options from clang to test
>         it here as
>         > > well.
>         >
>         >  I'm missing something probably - what's the benefit of that
>         over just using
>         >  -mllvm to pass this flag to llvm directly? (If I'm remembering
>         how this stuff
>         >  works)
>
>         I'm not sure - I was matching the -coverage-cfg-checksum and
>         -coverage-no-function-names-in-data CC1 options here. Maybe Nick has
>         thoughts?
>
>
>     LLVM flags are global state which will carry from one TU to the
>     next. Suppose you want to use clang as a library, and you want to
>     pass flags for one TU, then parse another TU with different flags.
>     It is an error to set the same LLVM flag twice, so that will
>     terminate your program. In general the -mllvm flags are bad news and
>     we should get rid of all of them and certainly not add new ones.
>
>
> I figured this flag was just as a utility for experiments - if it was
> for more than experiments I would imagine it'd need to come up to a
> frontend option, not just a -cc1 option, no?

It would be great to have a frontend option where you set what gcc 
version we want to emulate for gcov and it sets the right cc1 flags, but 
nobody has signed up for doing that yet.