Created attachment 8785770 [details] [diff] [review] bug-1298733-v1.patch Example run with local devices shows Interesting crashes. https://treeherder.allizom.org/#/jobs?repo=mozilla-inbound&revision=aef078d44124e895091e7341ae348fe3bb0c2255&filter-searchStr=autophone&group_state=expanded I think Mwglm is short enough that we can support it on production with our current inventory. Mwglc takes too long to support by default but we can support it on try. Thoughts on names/symbols etc?
Comment on attachment 8785770 [details] [diff] [review] bug-1298733-v1.patch Review of attachment 8785770 [details] [diff] [review]: ----------------------------------------------------------------- ::: configs/mochitests-webgl-conf-settings.ini @@ +25,5 @@ > +total_chunks = 10 > + > +[preferences] > +gfx.canvas.azure.backends=skia > +gfx.canvas.azure.accelerated=true I will let snorp comment on these prefs if they are proper, or if we have all of them. @@ +29,5 @@ > +gfx.canvas.azure.accelerated=true > + > +[treeherder] > +job_name = Autophone Mochitest WebGL conf > +job_symbol = Mwglc we assume WebGL == gl, on the tree, so could we just make this gl-c ? ::: configs/mochitests-webgl-mochitest-settings.ini @@ +29,5 @@ > +gfx.canvas.azure.accelerated=true > + > +[treeherder] > +job_name = Autophone Mochitest WebGL mochitest > +job_symbol = Mwglm I would like 'gl' as the symbol name
Attachment #8785770 - Flags: review?(jmaher) → review+
The prefs came from the Skia tests and may not be appropriate. snorp will know. The webgl-conf in treeherder appear as gl<chunk> but I think their actual symbols are Mgl<chunk>. They appear as gl<chunk> due to the way the Mochitests are combined in the output. We could make them gl<chunk> and they would appear similarly to the current webgl-conf tests. I don't actually see where the webgl-mochitest tests are being run any where. They should be different. Perhaps just glm.
I like 'gl' as the name too. The crashes we'll need to get fixed up, but I'm not sure about the other failures. It could be that those features are just not present on the hardware and we don't check for that? Input from Jeff here would be helpful.
Created attachment 8786720 [details] [diff] [review] bug-1298733-v2.patch -+job_symbol = Mwglc ++job_symbol = gl -+job_symbol = Mwglm ++job_symbol = glm A previous patch in bug 1298732 for using Treeherder to parse the logs didn't handle the marking of failing tests properly. A new local run with the new job symbols is in progress which better marks failing tests. The results are available at https://treeherder.allizom.org/#/jobs?repo=mozilla-central&revision=506facea63169a29e04eb140663da1730052db64&filter-searchStr=autophone&group_state=expanded&selectedJob=4409398 See bug 1298732 for more details on the log parsing issues. Carrying forward the r+
If it's different hardware, it'll have different failures. (and potentially different crashes)
Created attachment 8787242 [details] samsung gs3/nexus 6p webgl-crashes.txt Up to date local test run: https://treeherder.allizom.org/#/jobs?repo=mozilla-central&revision=b7f7ae14590aced450bb0b0469dfb38edd2c0ace&filter-searchStr=autophone&group_state=expanded
deployed 2016-09-04 ~13:40 PDT leave open until we run in production.
These are available on try and won't be available for CI until they are green and/or we have more devices for testing. Marking fixed.
Status: ASSIGNED → RESOLVED
Last Resolved: 9 months ago
Resolution: --- → FIXED
You need to log in before you can comment on or make changes to this bug.