It's really critical to switch to harfbuzz-ng because so many bugs are resolved with the switch.
Created attachment 178624 [details] Patch
Just enabled harfbuzz-ng in gyp without any C++ code change. The original Android-specific code in ComplexTextControllerHarfBuzz.cpp (depending on harfbuzz-old) is not needed in the new code path because we already have proper font fallback mechanism on in Skia for Android and used it in WebKit. Perhaps we should cleanup the obsolete code if no one uses harfbuzz-old.
Do we want this before the M25 branch point? It seems like a risky change to do on such short notice..
(In reply to comment #3) > Do we want this before the M25 branch point? It seems like a risky change to do on such short notice.. This is not urgent. I think we can land it after M25 branch point.
I doubt its particularly risky, given that linux has been on ng for several months iirc. But there is also no hurry in my view.
Comment on attachment 178624 [details] Patch Lgtm, but be sure to coordinate with peter wrt when this should be landed.
Is anyone still on old harfbuzz after this change? Can we delete the old harfbuzz code (there was some extra abstraction to handle both).
(In reply to comment #5) > I doubt its particularly risky, given that linux has been on ng for several months iirc. But there is also no hurry in my view. We used to have a few modifications in harfbuzz in regards to memory usage, switching Android to harfbuzz-ng just before the branch point can have unexpected side-effects unless it's really thoroughly tested. While it's not a particularly risky change, it's safer to do it in the M26 period. The amount of issues we have with M25 is rather high as it is. Landing this on Tuesday morning (assuming branch occurs on Monday) is perfectly fine with me. Thanks for doing this, Xianzhu!
Comment on attachment 178624 [details] Patch Clearing flags on attachment: 178624 Committed r138401: <http://trac.webkit.org/changeset/138401>
All reviewed patches have been landed. Closing bug.
Awesome! Thanks everyone.