clang: Suppress a -Wstring-conversion warnings in nspr, gyp edition.

clang: Suppress a -Wstring-conversion warnings in nspr, gyp edition.

It started firing in a few more cases recently.

BUG=82385
R=hans@chromium.org

Committed: https://src.chromium.org/viewvc/chrome?view=rev&revision=293109

[Nico, 2014-11-25 21:30:35]

thakis@chromium.org changed reviewers:
+ hans@chromium.org

[hans, 2014-11-25 21:31:57]

lgtm

[Nico, 2014-11-25 21:33:58]

Committed patchset #1 (id:1) manually as r293109 (presubmit successful).

[wtc, 2014-12-01 19:05:11]

wtc@chromium.org changed reviewers:
+ wtc@chromium.org

[wtc, 2014-12-01 19:05:12]

Patch set 1 LGTM.

I will fix this warning properly in NSPR upstream:
https://bugzilla.mozilla.org/show_bug.cgi?id=1106600

Since NSS is being replaced in Chromium soon, it is fine to just suppress this
compiler warning in Chromium.

https://codereview.chromium.org/762483003/diff/1/nss.gyp
File nss.gyp (right):

https://codereview.chromium.org/762483003/diff/1/nss.gyp#newcode257
nss.gyp:257: # nspr uses assert(!"foo") instead of assert(false && "foo").

PR_ASSERT(!"foo") can be replaced with PR_NOT_REACHED("foo"). I will fix these
in this way in the NSPR upstream.

PR_NOT_REACHED is defined as
http://mxr.mozilla.org/nspr/ident?i=PR_NOT_REACHED

The definition doesn't use the !"foo" construct.

Are you sure the compiler won't warn about assert(false && "foo")? I will need
to use that construct in one place, where the NSPR file defines its own
assertion macro.

[Nico, 2014-12-01 19:10:44]

On 2014/12/01 19:05:12, wtc wrote:
> Patch set 1 LGTM.
> 
> I will fix this warning properly in NSPR upstream:
> https://bugzilla.mozilla.org/show_bug.cgi?id=1106600
> 
> Since NSS is being replaced in Chromium soon, it is fine to just suppress this
> compiler warning in Chromium.
> 
> https://codereview.chromium.org/762483003/diff/1/nss.gyp
> File nss.gyp (right):
> 
> https://codereview.chromium.org/762483003/diff/1/nss.gyp#newcode257
> nss.gyp:257: # nspr uses assert(!"foo") instead of assert(false && "foo").
> 
> PR_ASSERT(!"foo") can be replaced with PR_NOT_REACHED("foo"). I will fix these
> in this way in the NSPR upstream.
> 
> PR_NOT_REACHED is defined as
> http://mxr.mozilla.org/nspr/ident?i=PR_NOT_REACHED
> 
> The definition doesn't use the !"foo" construct.
> 
> Are you sure the compiler won't warn about assert(false && "foo")? I will need
> to use that construct in one place, where the NSPR file defines its own
> assertion macro.

Yes, false && "foo" is explicitly not warned on by clang as that's considered
the common idiom for "fail an assertion with a descriptive string". This pattern
is used heavily in clang's code itself.