From a M1 log for a Linux opt build 6844 INFO TEST-END | /tests/content/base/test/test_bug338583.html | finished in 101694ms The entire M1 run takes 21 minutes, so the EventSource test takes 8% of the total test time to run (including setup and teardown). This is far too much time. Looking at the test I see lots of setTimeouts with intervals in the seconds (presumably? this dynamically calculated stress_factor obscures things) ...

it takes quite a bit time sure, but certain networking related things are hard to test without timeouts, at least with the testing framework we have now. Hard cases are for example testing that we get x messages, not x-1 or x+1. Another difficult case is to test if that we don't get any messages. Though, perhaps delayedServerEvents.sjs could test less.

Hmm. Actually 8% is too much. Perhaps the timeouts could be 10 times lower than as it is right now. I will work on that tomorrow and this weekend.

With this patch my machine took between 10 and 11 seconds to run the test.

Comment on attachment 537372 [details] [diff] [review] patch rs=me I'll push this to tryserver before pushing to m-c

test_bug338583.html only takes 10-12 seconds now on all OSs: https://tbpl.mozilla.org/php/getParsedLog.php?id=6420362&full=1 7573 INFO TEST-END | /tests/content/base/test/test_bug338583.html | finished in 10717ms https://tbpl.mozilla.org/php/getParsedLog.php?id=6420285&full=1 6857 INFO TEST-END | /tests/content/base/test/test_bug338583.html | finished in 10595ms https://tbpl.mozilla.org/php/getParsedLog.php?id=6420482&full=1 7573 INFO TEST-END | /tests/content/base/test/test_bug338583.html | finished in 10862ms https://tbpl.mozilla.org/php/getParsedLog.php?id=6420477&full=1 7573 INFO TEST-END | /tests/content/base/test/test_bug338583.html | finished in 11255ms