[DO NOT LAND] Toggle have_futex

[DO NOT LAND] Toggle have_futex

[Tom (Use chromium acct), 2017-02-23 19:52:29]

thomasanderson@google.com changed reviewers:
+ dpranke@chromium.org, primiano@chromium.org

[Tom (Use chromium acct), 2017-02-23 19:52:29]

+dpranke and primiano
Any idea why 'analyze' determined that no compile was necessary?

[Dirk Pranke, 2017-02-23 20:09:48]

On 2017/02/23 19:52:29, Tom Anderson wrote:
> +dpranke and primiano
> Any idea why 'analyze' determined that no compile was necessary?

Looks like the header is not listed in //base/allocator/BUILD.gn

[Primiano Tucci (use gerrit), 2017-02-23 21:11:38]

On 2017/02/23 20:09:48, Dirk Pranke wrote:
> On 2017/02/23 19:52:29, Tom Anderson wrote:
> > +dpranke and primiano
> > Any idea why 'analyze' determined that no compile was necessary?
> 
> Looks like the header is not listed in //base/allocator/BUILD.gn

Wait, isn't analyze.py using the dependency graph generated by the compiler?
depending on GN files seems really fragile, especially for headers which are
added only in one .gn file but they are referenced by various translation units
in several targets.

[Dirk Pranke, 2017-02-23 21:40:51]

On 2017/02/23 21:11:38, Primiano Tucci wrote:
> On 2017/02/23 20:09:48, Dirk Pranke wrote:
> > On 2017/02/23 19:52:29, Tom Anderson wrote:
> > > +dpranke and primiano
> > > Any idea why 'analyze' determined that no compile was necessary?
> > 
> > Looks like the header is not listed in //base/allocator/BUILD.gn
> 
> Wait, isn't analyze.py using the dependency graph generated by the compiler?
> depending on GN files seems really fragile, especially for headers which are
> added only in one .gn file but they are referenced by various translation
units
> in several targets.

No, we don't directly use the graph generated by the compiler, because in order
to get that graph we'd have to actually do the compile, which would defeat much
of the point. 

`gn check` parses c++ and header files to do a rough approximation of the
results, but it is only rough (again, to be fast).

There are discussions about how to sync things up w/ the dependency graphs from
the compilers but we haven't yet figured out a great way to do that that is both
up-to-date, too expensive, and maintainable.