deno test --watch should always print the full test output

[KnorpelSenf]

When running a suite of tests with deno test --watch ./tests/, all tests are run initially. As soon as a file is changed on disk, only the affected tests are re-run. This is great.

What is less great is that the output is now trimmed down to only the tests that were executed. This means that I lose the overview over all tests as soon as I touch one of them. I have to either cancel the watch mode and start it again, or manually save a bunch of unchanged test files to slowly reveal all output again.

Consider the following workflow:

1. Run all test in watch mode
2. See that two test files each have a failing test
3. Fix the first test, triggering a re-run
4. [Now I want to see all tests again to find out where the second test was failing]
5. Fix the second test
6. [Now I want to be sure that the entire test suite is green]

I suggest to keep the output from previous test runs in memory and to merge it with the new output. This would combine the best of both worlds: only affected tests are run, and the output of deno test --watch still reveals the results of the entire test suite.

[MKRhere]

As it is currently, when tests update, the report fails to give me the big picture of where all my tests are. It'd be cool if I can always see all test results. Failing that, if you want to keep output relevant to currently affected tests, at least keep the final counter overall, so I can glance at overall stats.

[bartlomieju]

This seems very reasonable. We might not have bandwidth to work on this for the next few weeks, but I'm more than happy to accept contributions.

[HasanAlrimawi]

@bartlomieju
Upon investigation,
I noticed that every test report is done by a thread with no persistence over the test results (results are generated and consumed on the same thread).
Then upon any change on any file, a new thread will be spawned to take responibility of reporting the same way. In short, it seems there isn't any mechanism to hold the tests records.

Approach I have in mind: How about passing through mutex wrapped struct that will be updated by every thread in order to keep overall statistics over all the files tested?

Is this an acceptable approach or is there a better approach that's maybe already used somewhere in the code base that you may suggest?

[marvinhagemeister]

+1 for moving reporting out of each individual thread to a central place. That would also allow us to print out test coverage information at the end and stuff like that.