Bug 95854 – [Chromium] Compositor build errors prevent rolling Chromium into WebKit

• Component: WebKit API
• Status: RESOLVED FIXED

[Peter Beverloo]

Example log:
http://build.webkit.org/builders/Chromium%20Linux%20Release/builds/58456/steps/compile-webkit/logs/stdio

In file included from ../../Source/WebKit/chromium/webkit/compositor_bindings/web_compositor_support_impl.cc:5:0:
../../Source/WebKit/chromium/webkit/compositor_bindings/web_compositor_support_impl.h:8:73: fatal error: third_party/WebKit/Source/Platform/chromium/public/WebLayer.h: No such file or directory

Include paths in web_compositor_support_impl.cc seem to be incorrect. Target range is 154695:154940, so I'm guessing r154819 is the issue here:
http://src.chromium.org/viewvc/chrome?view=rev&revision=154819

James, I don't know enough about this to quickly write up fixes. Could you confirm or correct the following assertions?

- There are two compositors, one in Chromium's cc/ directory, one in WebKit. The gyp "use_libcc_for_compositor" variable toggles this: 0 means WebKit, 1 means cc/.
- Chromium and WebKit currently both build with use_libcc_for_compositor=0. Are there bots which have use_libcc_for_compositor=1? Is there a bug or issue I can track for progress on this?

The canaries did not catch this error. The paths used on lines 21 to 31 of web_compositor_support_impl.cc cannot possibly work for WebKit, considering there is no such things as third_party/WebKit/ there.

Assuming we do need the "webkit_compositor_support" target itself, we need to change the include paths to work correctly under all scenarios. I'll prepare a patch for fixing this on the Chromium side..

[Peter Beverloo]

+Philippe, who needs a roll.

[Peter Beverloo]

I've uploaded a build fix here:
https://chromiumcodereview.appspot.com/10905100/

This fixes the issues with the WebKit build. It's a minor fix, so feel free to take over if there's a better way of doing it.

[James Robinson]

That fix is wrong.  We put third_party/WebKit/... into the gendir at build time so #include "third_party/WebKit/..." from chromium files should work.  I'll take a look at what's going on here - guessing some target doesn't have the correct gendir on the include path.

[James Robinson]

This is the rule in WebKit.gyp that's supposed to make things work:

                { # else: inside_chromium_build==0
                    'direct_dependent_settings': {
                        'include_dirs': [
                            '<(SHARED_INTERMEDIATE_DIR)/webkit', # in a chromium-inside-WebKit build, headers in the public WebKit API are copied beneath this directory so includes referencing third_party/WebKit work.
                        ],
                    },
                }],

but since webkit_compositor_support doesn't directly depend on the WebKit target, it doesn't pick this up.  One fix might be to just make it depend on WebKit.gyp:webkit - trying that now.  Another would be to duplicate this logic.

[James Robinson]

Possible fix here: https://chromiumcodereview.appspot.com/10918076. Gonna make sure trybots compile before landing but it appears to work locally (cr-mac).  ETA 20 minutes, feel free to land that if you're blocked on a roll and can't wait that long.

[Peter Beverloo]

(In reply to comment #5)
> Possible fix here: https://chromiumcodereview.appspot.com/10918076. Gonna make sure trybots compile before landing but it appears to work locally (cr-mac).  ETA 20 minutes, feel free to land that if you're blocked on a roll and can't wait that long.

SGTM, thank you James. Could you reply to the two assertions I made in comment 0?

[James Robinson]

(In reply to comment #0)
> - There are two compositors, one in Chromium's cc/ directory, one in WebKit. The gyp "use_libcc_for_compositor" variable toggles this: 0 means WebKit, 1 means cc/.

Yes (technically they're copies of each other, we're moving the WebKit one into cc/)

> - Chromium and WebKit currently both build with use_libcc_for_compositor=0. Are there bots which have use_libcc_for_compositor=1? Is there a bug or issue I can track for progress on this?

There's no bot.

[James Robinson]

Should be good as of https://src.chromium.org/viewvc/chrome?view=rev&revision=154974. Sorry for the mess.

[Peter Beverloo]

(In reply to comment #8)
> Should be good as of https://src.chromium.org/viewvc/chrome?view=rev&revision=154974. Sorry for the mess.

Thanks for fixing it! Any LKGR > 154974 should be good for a Chromium roll then, I'll take care of this tomorrow morning if no one else has by then :-).