Switch to BoringSSL.

Switch to BoringSSL.

This is a reland of r284079 which was reverted in r284248 for components build
issues. That, in turn, was a reland of r283813 which was reverted in r283845
because it broke WebRTC tests on Android. That, in turn, was a reland of
r283542 which was reverted in r283591 because it broke the WebView build.

This is a much larger change than its diff suggests. If it breaks
something, please revert first and ask questions later.

BUG=393317
R=agl@chromium.org, jam@chromium.org

Committed: https://src.chromium.org/viewvc/chrome?view=rev&revision=284729

[davidben, 2014-07-20 05:58:07]

This is a merger of the previous iteration, along with
https://codereview.chromium.org/404613002/
https://codereview.chromium.org/401943002/
https://codereview.chromium.org/398543003/
keystore_openssl.cc fixes to pick up the flag
Build BoringSSL with -Bsymbolic in the components build[*]
Update boringssl's BUILD.gn for components build changes

Unfortunately, https://codereview.chromium.org/401903004/ hasn't been picked up
by git yet, so it's missing try coverage with that. Hopefully that'll be sorted
out by Monday.

[*] This is for linux_chromium_clang_dbg which happens to build everything with
the components build, even though BoringSSL isn't used. Linux + x86_64 +
components otherwise hits this problem:

http://rt.openssl.org/Ticket/Display.html?id=2466&user=guest&pass=guest
https://sourceware.org/ml/binutils/2006-09/msg00149.html

Upstream's fix was
http://cvs.openssl.org/chngview?cn=20836
which got lost when the field moved to cpu-intel.c.

Mirroring that works as an ad-hoc fix, but I think -Bsymbolic is tidier. Making
all symbols overridable (what happens without -Bsymbolic) is kind of crazy and
we aren't -fvisibility=hidden clean. If we ever want to fully support a shared
object build, I think it'd be better to build with -fvisibility=hidden and
whitelist the exported API, rather than blacklist the symbols that cause us
problems.

[agl, 2014-07-20 12:29:15]

LGTM.

(Don't wait for me to LGTM everything - the only sane solution is to go nuts
with unreviewed changes.)

It does look like the git updater for the OpenSSL repo is bust - you might want
to poke chrome-infra about that today.

[davidben, 2014-07-21 16:44:46]

On 2014/07/20 12:29:15, agl wrote:
> LGTM.
> 
> (Don't wait for me to LGTM everything - the only sane solution is to go nuts
> with unreviewed changes.)
> 
> It does look like the git updater for the OpenSSL repo is bust - you might
want
> to poke chrome-infra about that today.

Hrmf. Apparently the OpenSSL repo is bust because of
https://code.google.com/p/chromium/issues/detail?id=395612. If that doesn't get
fixed in some reasonable time today, I'll probably try going without it. Not
aware of any bot configurations that care.

[jam, 2014-07-22 16:54:38]

lgtm for content and chrome

[davidben, 2014-07-22 18:20:49]

Committed patchset #12 manually as r284729.