JUnit output incomplete, truncated at ^2kB boundary

[allenap]

Running elm-test and redirecting to a file works as expected:

$ elm-test --report=junit > foo.xml
$ ls -l foo.xml
.rw-r--r-- gavin staff 789.9 KB Fri May 22 13:28:05 2020 foo.xml
                       ^^^^^^^^

Piping via another process results in a truncated output file:

$ elm-test --report=junit | cat > foo.xml
$ ls -l foo.xml
.rw-r--r-- gavin staff 64 KB Fri May 22 13:23:39 2020 foo.xml
                       ^^^^^

Observed on macOS (truncates at 64kB) and Linux (truncates at 128kB).

Just a guess, but I suspect there's a missing flush somewhere.

$ elm-test --version
0.19.1-revision3

[harrysarson]

I would expect that everything is flushed when the procress exits even if we forget to flush when we should?

[allenap]

There's some discussion of a similar bug here – and the root cause seems to be that Node does not flush process.stdout before exiting.

[harrysarson]

Ouch that is nasty paper cut in node. Here is a list of all the places the test-runner exits:

https://github.com/rtfeldman/node-test-runner/search?q=%22process.exit%22&unscoped_q=%22process.exit%22

PR that addsprocess.[stdout|stderr].flush() before each welcome!

[harrysarson]

Or there is the following recomendation here nodejs/node#6456

[process.stdout, process.stderr].forEach((s) => {
  s && s.isTTY && s._handle && s._handle.setBlocking &&
    s._handle.setBlocking(true)
})

[allenap]

[process.stdout, process.stderr].forEach((s) => {
  s && s.isTTY && s._handle && s._handle.setBlocking &&
  ...

Interesting that this checks for a TTY. This is what the set-blocking package does too, which is mentioned in node-test-runner's package-lock.json – though I can't find any uses of it.

Anyway, this issue is about behavior when writing to a pipe, not an interactive terminal, so I don't think that code snippet as is, or the set-blocking package, would resolve this bug.

One possible solution might be to call process.stdout._handle.setBlocking(true) regardless of whether we're connected to a TTY or not. The set-blocking package used to do this but changed its mind. The comments linked from there and other reading about this I've done lead me to believe that this situation in Node.js is a minefield, and far from resolved.

If calling setBlocking(true) regardless of TTY works, I suspect that might be a good solution right now for node-test-runner, on the basis that it's simple, not invasive, and easy to revert.

(Aside: should I refile this issue against node-test-runner? I always forget about the split between this package and that.)