1908744 - Update ANGLE to 9eadd12e33d65b8c9101ddef1d77d1c9a07e07bf (Chrome Beta 148.0.7778)

Status: RESOLVED FIXED

(Core :: Graphics, task, P1)

Comment 1

[Ashley Hale [:ahale]]

I'm going to attempt to update ANGLE to 7599 (very latest version) as we normally update to chromium beta's stable angle but this will land a bit late and there are some fixes after current stable angle that we want.

This is a first time for me updating ANGLE, and our documentation is hard to find and a little bit-rotted so I am likely to write new docs to put on firefox-source-docs based on my findings.

Comment 2

[Ashley Hale [:ahale]]

Note to self: https://bugzilla.mozilla.org/show_bug.cgi?id=2008997 has been applied to gecko but not mozilla/angle, will need to create the commit (with proper author).

Comment 3

[Jamie Nicol [:jnicol]]

I ended up doing a lot of the work for this to unblock myself for bug 2035532. It's not ready to land but should be a good starting point for someone to take over. Or I can resume when I get blocked again.

Firefox branch: https://github.com/jamienicol/firefox/commits/angle-update/
Angle branch: https://github.com/jamienicol/angle/commits/firefox-update/

The angle branch is branch firefox-129 of https://github.com/mozilla/angle/ rebased on to ANGLE branch chromium/7778. Though it's both missing a few of the most recent fixes from firefox-129, and the base is now few commits behind chromium/7778. (And 7778 is no longer the current beta branch). So it should be rebased again onto a newer upstream base, and the missing commits from firefox-129 should be cherry picked.

The patch stack is much shorter than before: the upstream cherry picks we carried can now be dropped, I squashed the "add missing ..." commits into the commits they were missing from to tidy it up, and the support for compiling with pre-C++20 can be dropped.

I got claude to do some of the conflict resolution where code had significantly changed, so these should be looked at carefully.

On the firefox side I think there are a few things to figure out:

• ShCompileOptions::enforcePackingRestrictions was removed upstream. See the angle bug for details, but essentially they don't believe this check is worthwhile at compile time (or translation time). I think they still enforce it at link time, but we'll only use ANGLE to link shaders on windows.
  • Chrome therefore skips these tests in "no passthrough" mode (which doesn't use ANGLE), but runs them in "passthrough" mode (which does use angle).
  • We need to decide whether to accept these tests as failing on non-Windows (i.e. on platforms that only use angle for shader translation only rather than for rendering), revert the patch and accept the maintainance burden of such, or add equivalent validation to our WebGL code.
• The immutable-tex-render-feedback CTS tests now fail due to this angle change. Angle thinks the CTS is wrong, and the tests are marked as failing in chromium too. So I think it's an easy call for us to do the same, but thought it was worth calling out
• ANGLE dropped support for GLSL version < 150 as a shader translator output target. Beacuse ANGLE no longer support GL versions < 3.2. This will mean we can no longer translate shaders for GL context versions < 3.2, and therefore cannot support webgl. This affects the translator, i.e. all platforms not just windows.
  • So we either need to
    a) block webgl versions which don't support that GLSL/GL version
    b) revert this patch and accept the maintainence burden
    c) Or see if we can get angle to revert the patch and continue supporting these targets
  • For what it's worth, I checked telemetry to see what percentage of our users have OpenGL < 3.2. We don't expose the version directly but using the adapter id as a proxy claude estimated about 1.2%. Which I think makes a) unviable.

Comment 4

[Jamie Nicol [:jnicol]]

ANGLE dropped support for GLSL version < 150 as a shader translator output target

To summarise some offline discussions: Ashley spoke to ANGLE folks and they dropped support because their number of affected users was negligible. Jeff found that there exists some hardware that does not support desktop GL 3.2, but probably does support GLES 2.0. We should therefore attempt to give these users WebGL 1 on top of GLES 2.

So the plan was to accept dropping support for GLSL versions < 150, but ensure we attempt to create a GLES 2 context rather than a desktop GL < 3.2

Comment 5

[Erich Gubler [:ErichDonGubler] (he/him)]

I've created the firefox-153 branch in mozilla/angle to track the patches being accumulated from the current base.

Comment 6

[Erich Gubler [:ErichDonGubler] (he/him)]

Attachment: Bug 1908744 - build(angle): enforce a `git log --date` for consistency r=#gfx-reviewers

I moved to the Netherlands recently from Florida. To my utter non-surprise, the system locale changed how things were rendered! This is a problem for re-vendoring ANGLE specifically, because the artifacts produced in moz.yaml and cherry_picks.txt are supposed to be consistent between the users who run them. Ugh!

Squash some system-dependent behavior by passing a consistent --date=format:… to all invocations of git log in the vendor_from_git Python module, and revendor ANGLE from the same commit. The only change this applies is a leading zero to some commit timestamps; otherwise, there are no changes to imported ANGLE code. This was necessary because there's no locale-independent way of asking Git to print dates without a leading space (i.e., %e) or leading zero (i.e., %d). So, we opt for the latter.

I doubt we've squashed everything system-dependent here, but this was enough to get things back to "normal" on my machine.

Comment 7

[Erich Gubler [:ErichDonGubler] (he/him)]

Attachment: Bug 1908744 - chore(angle): update `KNOWN_ON_CI` renderer list with new ANGLE value r=#gfx-reviewers

ANGLE commit aaeeea0c42a877535428b146c7fd342ec19f0529 started adding the Device ID to the renderer string. We must update the KNOWN_ON_CI renderer list to reflect the new format. We've only observed the last item in the list, so only add it; if we need to add more later, we can do so in a similar manner.

Comment 8

[Erich Gubler [:ErichDonGubler] (he/him)]

Attachment: Bug 1908744, part 5 - chore(angle): update base to 9eadd12e33d65b8c9101ddef1d77d1c9a07e07bf r=#gfx-reviewers

It's been years in the making! 😅 I return to revendoring ANGLE with
tons of support from @jnicol, who has done much (most?) of the work
associated with this patch.

There are some important call-outs in this patch; this is mostly content written by @jnicol, which I've glued together into a single commit.

Overall scope

It's been years since we revendored ANGLE. There are a huge number of changes, and that makes reviewing this…well, prohibitively difficult, to say the least! My suggestion is to not try to review the entire set of changes, much like we don't force a reviewers to deal with all source for code in, say, third_party/ (which we eventually hope to move our checkout to).

update-angle.py changes

The new library angle_common_shader_state is now used by both the translator and libGLESv2. We therefore add it to gfx/angle/moz.build's DIRS like we do for angle_common, and use already_added_dirs to prevent it being added multiple times.

Desktop GL front-end support has been removed upstream (in commit b16d105fc63a7365d6809d5033cb1b2172d8d311) so no longer needs to be explicitly disabled.

Conversely, there now exists a wgpu/Dawn backend which must be explicitly disabled.

The angle_enable_apple_translator_workarounds arg has been removed (in commit 20b6624a88a599fec92cceaf7d80b0e1141449cf). We must instead manually ensure the required sources are vendored and added to the translator's sources for macOS builds. In the future the need for hacks like this will be obviated by switching to gn_processor.py to generate our builds scripts in bug 2035532.

Due to update-angle.py flattening GN source_sets' sources into the libraries which depend on them, some sources are compiled multiple times in different libraries which are then linked together, causing linker errors. We manually remove problematic sources from one of their libraries to avoid this. Again, this will be fixed in a more robust manner by bug 2035532 (as gn_processor.py emits source_sets as static libraries).

Lastly, the library //third_party/llvm-build/Release+Asserts/lib/clang/23/lib/windows/clang_rt.builtins-x86_64.lib is being pulled in as a dependency for some reason, so we filter out any dependencies beginning with //. And synchronization is now brought in as a genuine dependency on Windows, but we must avoid adding this to OS_LIBS on other platforms. For now, we only ever add it to OS_LIBS on Windows, as we know that other platforms have no OS lib dependencies. Once again, these issues will be fixed properly in bug 2035532.

Test expectations

immutable-tex-render-feedback tests now fail due to an upstream change where they believe the CTS is incorrect.

uniform-packing-restrictions tests now fail on platforms where ANGLE is only used for shader translation, due to ANGLE no longer enforcing these rules at compile time.

EXT_depth_clamp tests now pass on Windows, and vector-matrix-constructor-scalarization tests now pass on macOS, due to upstream fixes.

Comment 9

[Erich Gubler [:ErichDonGubler] (he/him)]

:jnicol and I exhaustively went over every downstream commit we had in firefox-129, and determined what we need to do for each of them to ensure we don't lose anything important. Things generally fell into these buckets:

1. A fix is already in the new ANGLE base, no action item needed (beyond maybe dropping the commit because it conflicts instead of empties out).
2. The fix needs to be carried forward because it's not yet in ANGLE upstream (at the point we're revendoring from, anyway).

I have updated the firefox-153 branch to accommodate all of the known action items except a tricky forward-porting reversion of jamienicol/firefox:092bd3b41b224d823eb601a3e47e2ea3999fb0e4 (ShCompileOptions being removed) and at least some work to correspondingly revert it in ANGLE, which I plan to work on today. I think this is close enough that we can start testing in CI, and maybe even a review process for the stack.

Here's a roadmap of what's left:

1. Figure out the forward-port reversion of ShCompileOptions and consume it, adjusting test expectations accordingly.
2. Do some testing to figure out what regressions might be present.
3. Complete a review process.

I think (1) and (2) can happen independently, and (3) is soft-dependent on (2) so a reviewed patch doesn't get completely invalidated.

Comment 10

[Norisz Fay [:noriszfay]]

https://hg.mozilla.org/mozilla-central/rev/774fb825cf6e

Comment 11

[Pulsebot]

Pushed by [e-mail]:
https://github.com/mozilla-firefox/firefox/commit/f67420a758a2
https://hg.mozilla.org/integration/autoland/rev/f2a891d77669
chore(angle): update `KNOWN_ON_CI` renderer list with new ANGLE value r=gfx-reviewers,aosmond

Comment 12

[amarc]

https://hg.mozilla.org/mozilla-central/rev/f2a891d77669

Comment 13

[Erich Gubler [:ErichDonGubler] (he/him)]

Attachment: Bug 1908744, part 1 - chore(angle): stop using removed `ShCompileOptions.variables` config r=#gfx-reviewers

options stopped having any effect in upstream angle commit 02e7f967597a2d706517c456739f6aecd4828a52, when its behaviour was turned on by default. The flag was later removed altogether in commit 37698dcf13dd61a3baac496eebc6bfd151df352f.

Comment 14

[Pulsebot]

Pushed by [e-mail]:
https://github.com/mozilla-firefox/firefox/commit/261d2a25bfbd
https://hg.mozilla.org/integration/autoland/rev/2cd63d2bf914
chore(angle): remove unused `ShCompileOptions.variables` config r=gfx-reviewers,jnicol

Comment 15

[Pulsebot]

Pushed by [e-mail]:
https://github.com/mozilla-firefox/firefox/commit/d70684b03aa4
https://hg.mozilla.org/integration/autoland/rev/2b246d425a48
Revert "Bug 1908744 - chore(angle): remove unused `ShCompileOptions.variables` config r=gfx-reviewers,jnicol" for causing failures @test_conformance__attribs__gl-bindAttribLocation-aliasing.html.

Comment 16

[agoloman]

Backed out for causing failures @test_conformance__attribs__gl-bindAttribLocation-aliasing.html.

• Backout link
• Push with failures
• Failure Log

Comment 17

[Jamie Nicol [:jnicol]]

Erich, I think you have to land that one along with the update - the variable still has an effect in the current version of ANGLE we have vendored

Comment 18

[Erich Gubler [:ErichDonGubler] (he/him)]

Attachment: Bug 1908744, part 3 - chore(angle): update `update-angle.py` r=#gfx-reviewers

(NOTE: Most of this patch's development was done by @jnicol. I stand behind it as a patch I understand, but want to recognize his efforts as most of its material.)

The new library angle_common_shader_state is now used by both the translator and libGLESv2. We therefore add it to gfx/angle/moz.build's DIRS like we do for angle_common, and use already_added_dirs to prevent it being added multiple times.

Desktop GL front-end support has been removed upstream (in commit b16d105fc63a7365d6809d5033cb1b2172d8d311) so no longer needs to be explicitly disabled.

Conversely, there now exists a wgpu/Dawn backend which must be explicitly disabled.

The angle_enable_apple_translator_workarounds arg has been removed (in commit 20b6624a88a599fec92cceaf7d80b0e1141449cf). We must instead manually ensure the required sources are vendored and added to the translator's sources for macOS builds. In the future the need for hacks like this will be obviated by switching to gn_processor.py to generate our builds scripts in bug 2035532.

Due to update-angle.py flattening GN source_sets' sources into the libraries which depend on them, some sources are compiled multiple times in different libraries which are then linked together, causing linker errors. We manually remove problematic sources from one of their libraries to avoid this. Again, this will be fixed in a more robust manner by bug 2035532 (as gn_processor.py emits source_sets as static libraries).

Lastly, the library //third_party/llvm-build/Release+Asserts/lib/clang/23/lib/windows/clang_rt.builtins-x86_64.lib is being pulled in as a dependency for some reason, so we filter out any dependencies beginning with //. And synchronization is now brought in as a genuine dependency on Windows, but we must avoid adding this to OS_LIBS on other platforms. For now, we only ever add it to OS_LIBS on Windows, as we know that other platforms have no OS lib dependencies. Once again, these issues will be fixed properly in bug 2035532.

Comment 19

[Erich Gubler [:ErichDonGubler] (he/him)]

Attachment: Bug 1908744, part 4 - test(angle): update expectations r=#gfx-reviewers

(NOTE: Most of this patch's development was done by @jnicol. I stand behind it as a patch I understand, but want to recognize his efforts as most of its material.)

• immutable-tex-render-feedback tests now fail due to an upstream change where they believe the CTS is incorrect.
• EXT_depth_clamp tests now pass on Windows, and vector-matrix-constructor-scalarization tests now pass on macOS, due to upstream fixes.

Comment 20

[Erich Gubler [:ErichDonGubler] (he/him)]

:jkratzer: Since ANGLE is a sec-sensitive component, I wanted to ping you about this patch stack being ready enough to maybe point some auditing resources at early. Did you want to set something up before landing? Otherwise, I think this is basically ready to land after a typical review process.

Comment 21

[Erich Gubler [:ErichDonGubler] (he/him)]

Attachment: Bug 1908744, part 2 - chore(angle): migrate to reworked `Max{,Total}PrivateVariableSizeInBytes` limits r=#gfx-reviewers

One of the Mozilla-specific patches we had prior to this patch sequence exposed private variable size limits as user-configurable. At that time, the single limit member MaxPrivateVariableSizeInBytes enforced both individual and total private variable sizes. This wasn't necessarily the most intelligent check, but it did its job for us.

Upstream, two changes happened that we needed to account for with a new revision of the patch. The MaxPrivateVariableSizeInBytes limit was split into Max{,Total}PrivateVariableSizeInBytes, splitting out the individual and total size checks into two separate knobs. The former also requires opt-in with ShCompileOptions.rejectWebglShadersWithLargeVariables. Begin using both.

Comment 22

[Erich Gubler [:ErichDonGubler] (he/him)]

Okay, this patch stack is ready to go. Jamie and I have gone to some pretty significant to make sure we're not breaking anything important with the new rebase. So, I'm planning on landing this ASAP.

🫡 Thanks for the patience, everyone!

Comment 23

[Pulsebot]

Pushed by [e-mail]:
https://github.com/mozilla-firefox/firefox/commit/4f95d15c6332
https://hg.mozilla.org/integration/autoland/rev/cefaacd164b1
part 1 - chore(angle): stop using removed `ShCompileOptions.variables` config r=gfx-reviewers,jnicol
https://github.com/mozilla-firefox/firefox/commit/f45cc36d8576
https://hg.mozilla.org/integration/autoland/rev/f9501d3298de
part 2 - chore(angle): migrate to reworked `Max{,Total}PrivateVariableSizeInBytes` limits r=gfx-reviewers,jnicol
https://github.com/mozilla-firefox/firefox/commit/b7050d38b8ed
https://hg.mozilla.org/integration/autoland/rev/f57f4eb32ba3
part 3 - chore(angle): update `update-angle.py` r=gfx-reviewers,nical
https://github.com/mozilla-firefox/firefox/commit/828f6e7efe56
https://hg.mozilla.org/integration/autoland/rev/7019c4c5b85d
part 4 - test(angle): update expectations r=gfx-reviewers,nical
https://github.com/mozilla-firefox/firefox/commit/6e035dc7d652
https://hg.mozilla.org/integration/autoland/rev/4c67671fda1b
part 5 - chore(angle): update base to 9eadd12e33d65b8c9101ddef1d77d1c9a07e07bf r=gfx-reviewers,nical

Comment 24

[Pulsebot]

Pushed by [e-mail]:
https://github.com/mozilla-firefox/firefox/commit/1d312a19cc2e
https://hg.mozilla.org/integration/autoland/rev/75dc405e97bb
Revert "Bug 1908744, part 5 - chore(angle): update base to 9eadd12e33d65b8c9101ddef1d77d1c9a07e07bf r=gfx-reviewers,nical" for causing hazard failures.

Comment 25

[Cosmin Sabou [:CosminS]]

Backed out for causing hazard failures.

Push with failures

Failure log

Backout link

Comment 26

[Erich Gubler [:ErichDonGubler] (he/him)]

I…hadn't heard of hazard testing before! Time to learn. /me rolls up sleeves

Comment 27

[Cosmin Sabou [:CosminS]]

.

Comment 28

[Jamie Nicol [:jnicol]]

I think only partially backing this out might itself cause other failures. See comment 16 where attempting to land just the first patch by itself resulted in a backout. Should we back the whole thing out?

Comment 29

[Cosmin Sabou [:CosminS]]

It backed out the whole stack: https://hg-edge.mozilla.org/integration/autoland/rev/75dc405e97bb9f032dc94dabcd1358288e827047
Please ignore comment 27 as Bugherder should have done.

Comment 30

[Erich Gubler [:ErichDonGubler] (he/him)]

Attachment: Bug 1908744 - chore(hazards): only rewrite `shared_ptr<T>` dtors when `T::~T()` exists r=sfink

With the current state of the root analysis in hazards testing, two cases in patches for bug 1908744 fail:

• std::__future_base::_State_baseV2, whose destructor is defined out-of-line in libstdc++.
• angle::Closure, a polymorphic abstract base. std::make_shared is only used for its subclasses; the deleter runs the derived destructor, and the base destructor is never used directly.

Instead of asserting, only perform the rewrite when the synthesized destructor actually exists in the body database. When it does not, return nothing so the ~shared_ptr() call is analyzed normally.

Comment 31

[Phabricator Automation]

Comment on attachment 9604183 [details]
Bug 1908744 - chore(hazards): only rewrite shared_ptr<T> dtors when T::~T() exists r=sfink

Revision D310296 was moved to bug 2052331. Setting attachment 9604183 [details] to obsolete.

Comment 32

[Erich Gubler [:ErichDonGubler] (he/him)]

Okay, so, I've filed bug 2052331 to handle what seems like the only real CI failure with autoland checks. That's a neat bit of progress! Hopefully :sfink can help us get unblocked here soon. 🤞🏻

Comment 33

[Pulsebot]

Pushed by [e-mail]:
https://github.com/mozilla-firefox/firefox/commit/3670c00a7607
https://hg.mozilla.org/integration/autoland/rev/d3316cb73818
part 1 - chore(angle): stop using removed `ShCompileOptions.variables` config r=gfx-reviewers,jnicol
https://github.com/mozilla-firefox/firefox/commit/fd1892c99921
https://hg.mozilla.org/integration/autoland/rev/0a80bc7cfeb2
part 2 - chore(angle): migrate to reworked `Max{,Total}PrivateVariableSizeInBytes` limits r=gfx-reviewers,jnicol
https://github.com/mozilla-firefox/firefox/commit/33dd45304318
https://hg.mozilla.org/integration/autoland/rev/e88cd835b4bb
part 3 - chore(angle): update `update-angle.py` r=gfx-reviewers,nical
https://github.com/mozilla-firefox/firefox/commit/b11cc9a596f4
https://hg.mozilla.org/integration/autoland/rev/7b03fdad9cf3
part 4 - test(angle): update expectations r=gfx-reviewers,nical
https://github.com/mozilla-firefox/firefox/commit/2fe9f8877008
https://hg.mozilla.org/integration/autoland/rev/22c80afa7078
part 5 - chore(angle): update base to 9eadd12e33d65b8c9101ddef1d77d1c9a07e07bf r=gfx-reviewers,nical

Comment 34

[Pulsebot]

Pushed by [e-mail]:
https://github.com/mozilla-firefox/firefox/commit/a3d2605e3c70
https://hg.mozilla.org/integration/autoland/rev/0b736d13634f
Revert "Bug 1908744, Bug 2052331, part 5 - chore(angle): update base to 9eadd12e33d65b8c9101ddef1d77d1c9a07e07bf r=gfx-reviewers,nical" for causing Hazard Bustages.

Comment 35

[Atila Butkovits]

Backed out for causing Hazard Bustages.

Backout link: https://hg.mozilla.org/integration/autoland/rev/0b736d13634f

Push with failures: https://treeherder.mozilla.org/jobs?repo=autoland&resultStatus=testfailed%2Cbusted%2Cexception%2Cretry%2Cusercancel&revision=22c80afa707823d64937df0492e91173db62506d&searchStr=Linux%2Cdebug%2Chazard-linux64-haz%2Fdebug%2CH

Failure log: https://treeherder.mozilla.org/logviewer?job_id=577715894&repo=autoland&task=RALzl4q6Q3OB2IrGdbTifQ.0&lineNumber=9964

Comment 36

[pstanciu]

https://hg.mozilla.org/mozilla-central/rev/d3316cb73818
https://hg.mozilla.org/mozilla-central/rev/0a80bc7cfeb2
https://hg.mozilla.org/mozilla-central/rev/e88cd835b4bb
https://hg.mozilla.org/mozilla-central/rev/7b03fdad9cf3
https://hg.mozilla.org/mozilla-central/rev/22c80afa7078

Comment 37

[Takanori MATSUURA]

(In reply to pstanciu from comment #36)

https://hg.mozilla.org/mozilla-central/rev/d3316cb73818
https://hg.mozilla.org/mozilla-central/rev/0a80bc7cfeb2
https://hg.mozilla.org/mozilla-central/rev/e88cd835b4bb
https://hg.mozilla.org/mozilla-central/rev/7b03fdad9cf3
https://hg.mozilla.org/mozilla-central/rev/22c80afa7078

Backed out by https://hg-edge.mozilla.org/mozilla-central/rev/0b736d13634f.

Comment 38

[Pulsebot]

Pushed by [e-mail]:
https://github.com/mozilla-firefox/firefox/commit/c929fb16090d
https://hg.mozilla.org/integration/autoland/rev/b7131ffd2c21
part 1 - chore(angle): stop using removed `ShCompileOptions.variables` config r=gfx-reviewers,jnicol
https://github.com/mozilla-firefox/firefox/commit/fc8270fb0217
https://hg.mozilla.org/integration/autoland/rev/ab8aceebce73
part 2 - chore(angle): migrate to reworked `Max{,Total}PrivateVariableSizeInBytes` limits r=gfx-reviewers,jnicol
https://github.com/mozilla-firefox/firefox/commit/b0c7462fb774
https://hg.mozilla.org/integration/autoland/rev/e39ffef9d882
part 3 - chore(angle): update `update-angle.py` r=gfx-reviewers,nical
https://github.com/mozilla-firefox/firefox/commit/f2dfc2255c9f
https://hg.mozilla.org/integration/autoland/rev/53af7abe73ed
part 4 - test(angle): update expectations r=gfx-reviewers,nical
https://github.com/mozilla-firefox/firefox/commit/6a2def449e8b
https://hg.mozilla.org/integration/autoland/rev/16134e4ac77c
part 5 - chore(angle): update base to 9eadd12e33d65b8c9101ddef1d77d1c9a07e07bf r=gfx-reviewers,nical

Comment 39

[Atila Butkovits]

https://hg.mozilla.org/mozilla-central/rev/b7131ffd2c21
https://hg.mozilla.org/mozilla-central/rev/ab8aceebce73
https://hg.mozilla.org/mozilla-central/rev/e39ffef9d882
https://hg.mozilla.org/mozilla-central/rev/53af7abe73ed
https://hg.mozilla.org/mozilla-central/rev/16134e4ac77c

Comment 40

[Erich Gubler [:ErichDonGubler] (he/him)]

Attachment: Bug 1908744 - chore: touch `CLOBBER` for ANGLE update r=#gfx-reviewers

Comment 41

[Pulsebot]

Pushed by [e-mail]:
https://github.com/mozilla-firefox/firefox/commit/8bd6171f7241
https://hg.mozilla.org/integration/autoland/rev/3e15fb059117
chore: touch `CLOBBER` for ANGLE update r=gfx-reviewers,jnicol DONTBUILD

Comment 42

[Sandor Molnar[:smolnar]]

https://hg.mozilla.org/mozilla-central/rev/3e15fb059117