This bug is to investigate why files under 'devtools/' are being covered intermittently. This occurs at the level of the raw JSVM data so this is probably a bug in how JSVM coverage is collected, or generated. So far, we get coverage for them 25% of the time. I've attached a JSON containing some differences between the tasks: JSDCOV: https://treeherder.mozilla.org/#/jobs?repo=try&selectedJob=203191231&revision=22abb99d81005498bf86ec6cd896fb1c5e6c8fcf JSVM (this job specifically): https://treeherder.mozilla.org/#/jobs?repo=try&selectedJob=203191286&revision=f90b8cdb2cefb9e39f37db54415645df0055ae69 One of the 4 JSVM jobs (in the JSVM push above), contains coverage for nearly all 'devtools' files, making it almost identical to JSDCOV. The biggest differences were from 'testing/*' files being covered in JSDCOV but not JSVM - which we should be able to ignore. This is the job which is very similar: https://treeherder.mozilla.org/#/jobs?repo=try&selectedJob=201089393&revision=f90b8cdb2cefb9e39f37db54415645df0055ae69

Adding this as a dependency as it's likely that these files are only missing because of this error.

Greg, could you try to redo this analysis using the workaround suggested in bug 1496768 comment 7?

We can close this bug as it was entirely related to bug 1496768 and fixed with the workaround in bug 1496768 comment 7. Here is an example of a good run when `--rebuild` is used, showing that each chunk has JSVM data: https://treeherder.mozilla.org/#/jobs?repo=try&selectedJob=208213201&revision=61ca7efce095dc8dc1efe65c19cab166084d83e4 Want me to double-check JSDCOV vs JSVM differences? I have a feeling that it will result in the same result where JSVM and JSDCOV data do not differ (like the chunk that was good in the description comment).

I guess we can close this!