ScottFreeCode
commented
7 years ago I've been working on this in the multi-coverage branch.
Getting Coveralls to take reports from multiple build jobs in the same build turned out to be way easier than I thought. I wasted so many hours looking for a more complex solution because I didn't initially realize I could just apply COVERAGE=true to the whole build. That alone should increase the coverage somewhat as it will show branches taken in different Node versions; would be nice to get more such branches covered though (Windows, whatever other terminals we're handling).
After that, I looked into adding coverage to the browser runs. Here's what I found out.
karma-coveragecan instrument the code and report on coverage for instrumented code, however...karma-browserifydoesn't look at the results of other preprocessors, therefore...browserify-istanbulshould be used to makekarma-browserifyinstrument the code, which can then be reported upon bykarma-coverage.browserify-istanbulmysteriously doesn't run on anything going into the bundle that isn't in Karma'sfileslist and given['browserify']on Karma'spreprocessorslist, so we have to add the relevant patterns to those lists even if they shouldn't be served directly.- After I got all that figured it, it finally worked! ...Until I tried to clean up the other false starts I'd had, use NYC as described below, and consolidate the code behind
if (env.COVERAGE); then something changed and caused the current "bundle error" problems -- but I haven't been able to figure out what. - It is possible to override the coverage implementation for both
browserify-istanbulandkarma-coverage; this can be used to makebrowserify-istanbuluse the same version of Istanbul askarma-coverage, or to make both use the same version of Istanbul as our Node CLI NYC. (This seems to work ok except for the above-mentioned issue, and removing it does not seem to resolve the above-mentioned issue.)
Feel free to pull it down locally and experiment to confirm or deny any of the above and try to fix the bundle errors.
We have coverage numbers, but only in Node.js. I'd like to see the coverage numbers against tests run with Karma.
Gathering coverage from Karma seems fraught with peril.
I would entertain other suggestions to improve how we run browser-based tests. My only requirement is that the tests run in a strict browser-like environment--
jsdomshenanigans or Electron won't do.