Turn off nacl_bootstrap_helper when building for code coverage.

Turn off nacl_bootstrap_helper when building for code coverage.

BUG=None
TEST=None
R=bradchen@google.com
Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=109908

[Brad Chen, 2011-11-13 18:51:54]

LGTM

Can you help me understand the nature of the failure? I'd like to know if this
flakiness is also something that could induce failures in the field.

[bradn, 2011-11-14 17:13:33]

So actually its just failing due to -lgcov not being on the link line:
http://chromesshgw.corp.google.com/i/chromium.fyi/builders/Linux%20Coverage%2...

Would you rather I just added that to the link line conditionally?
I guess I wasn't sure how much sense building the helper with gcov made in
general? My understanding of how it's run is a little thin, do I understand
correctly that it won't have file access to emit gcov data?

[Roland McGrath, 2011-11-14 17:18:28]

On 2011/11/14 17:13:33, bradn wrote:
> So actually its just failing due to -lgcov not being on the link line:
>
http://chromesshgw.corp.google.com/i/chromium.fyi/builders/Linux%2520Coverage...

This link is stale.

> Would you rather I just added that to the link line conditionally?
> I guess I wasn't sure how much sense building the helper with gcov made in
> general? My understanding of how it's run is a little thin, do I understand
> correctly that it won't have file access to emit gcov data?

nacl_helper is a child of the zygote, running inside sandboxes so it indeed
has no direct filesystem access.

[bradn, 2011-11-14 17:22:17]

Fixed
link<http://chromesshgw.corp.google.com/i/chromium.fyi/builders/Linux%20Coverage%20%28dbg%29/builds/1028/steps/compile/logs/stdio>

Ok, sounds like disable is the way to go then.
I'll wait a little and commit if I don't hear anything to the contrary.

-BradN

On Mon, Nov 14, 2011 at 9:18 AM, <mcgrathr@chromium.org> wrote:

> On 2011/11/14 17:13:33, bradn wrote:
>
>> So actually its just failing due to -lgcov not being on the link line:
>>
>
> http://chromesshgw.corp.**google.com/i/chromium.fyi/**
> builders/Linux%2520Coverage%**2520%2528dbg%2529/builds/1028/**
>
steps/compile/logs/stdio<http://chromesshgw.corp.google.com/i/chromium.fyi/builders/Linux%2520Coverage%2520%2528dbg%2529/builds/1028/steps/compile/logs/stdio>
>
> This link is stale.
>
>
>  Would you rather I just added that to the link line conditionally?
>> I guess I wasn't sure how much sense building the helper with gcov made in
>> general? My understanding of how it's run is a little thin, do I
>> understand
>> correctly that it won't have file access to emit gcov data?
>>
>
> nacl_helper is a child of the zygote, running inside sandboxes so it indeed
> has no direct filesystem access.
>
>
>
http://codereview.chromium.**org/8538020/<http://codereview.chromium.org/8538...
>

[Use chromium.org instead, 2011-11-14 17:25:39]

nacl_helper_bootstrap should not be compiled with coverage options or
linked with any extra libraries.  It's a very special case, because it is
built to be very minimal.  It probably makes sense just to give it a fixed
special set of cflags regardless of the configuration.

That doesn't say anything about nacl_helper itself.  It probably makes as
much sense for that to be built in gcov mode as does building renderer code
that way.  How do other processes that are normally sandboxed get adequate
file access to write gcov output files?  If they're all just run without
the sandbox for gcov runs, that should be fine for nacl_helper too.

[bradn, 2011-11-14 17:28:06]

Most of it is not even run for gcov mode.

-BradN

On Mon, Nov 14, 2011 at 9:25 AM, Roland McGrath <mcgrathr@google.com> wrote:

> nacl_helper_bootstrap should not be compiled with coverage options or
> linked with any extra libraries.  It's a very special case, because it is
> built to be very minimal.  It probably makes sense just to give it a fixed
> special set of cflags regardless of the configuration.
>
> That doesn't say anything about nacl_helper itself.  It probably makes as
> much sense for that to be built in gcov mode as does building renderer code
> that way.  How do other processes that are normally sandboxed get adequate
> file access to write gcov output files?  If they're all just run without
> the sandbox for gcov runs, that should be fine for nacl_helper too.
>

[commit-bot: I haz the power, 2011-11-14 17:41:58]

No LGTM from valid reviewers yet.