jgraham: can you look at web-platform-tests/wpt #27699 once more?

It's the one remaining thing of a spec change that has already happened... I wish it wasn't merged until the tests are also merged, makes keeping track of things hard

annevk: Sure

ta

jgraham: so I am realizing that step_wait and friends don't actually offer functionality of doing something repeatedly

annevk: I'm not sure I understand. Do you have an example of what you can't do?

jgraham: I basically want to poll a URL until I get a certain result

Is the problem that the condition function is then async and returns a promise?

jgraham: I actually didn't dive into it, but I suspect so, I'll look at the code to double check

jgraham: yup, just if (cond()) which will be true for a promise...

jgraham: I guess I could workaround it sorta with some additional state variables to not duplicate fetches

Yeah, it shoud be possible to handle in user code without too much difficulty. But it seems like a rough edge in the harness for sure. DOes something like paste.mozilla.org/Mk96cHpU work? That's untested…

jgraham: it does not and I don't get errors either

jgraham: if I move the handle_result operation to be earlier it works

jgraham: there does seem to be a problem with this logic in that if the promise takes a long time the whole timeline is vastly extended

annevk: Right, it would need to actually compute the time that's passed since it started in order to make the timing accurate

Which I guess it can do if necessary, but it does add more DOM requirements on testharness.js

jgraham: although to be fair we don't care about this for expensive sync execution either

Yeah, the design assumes that the condition function is trivial in order to avoid depending on either Date() or performance.now()

jgraham: it would really help if the hidden assumptions behind testharness.js were documented

jgraham: it's not entirely clear to me for instance why let/const, arrow functions, etc. cannot be used

annevk: Honestly they probably can now

Previously there was a concern about testing older browsers

Now the only case where that might be relevant is the WMAS thing that tests TV browsers

(and also these things are not as new as they used to be)

(there is also code in testharness.js that deals with some postMessage bugs in old webkit, so there's lots of cleanup possible)

(but it's a high-risk/low-reward project since you can accidentially break things and at the end you don't have a great deal to show for it)