Josh Matthews

I set a breakpoint on ipc_channel::ipc::channel and looked at backtraces for a while. We still create an IPC channel per FetchCanceller, which is one per request, so we might want...

The panic is gone!

Nevermind, it's still happening.

There's one channel per image fetch (via `generate_cache_listener_for_element`), which appears to be the largest source of file descriptor exhaustion based on looking at backtraces from ipc_channel::ipc::channel(). Then each of those...

The panic is now intermittent, at least, rather than consistent like it used to be. Progress!

Hmm, on a different macOS machine I now see: ``` (:50386): GLib-ERROR **: 16:57:12.063: Creating pipes for GWakeup: Too many open files Servo was terminated by signal 5 ``` Edit:...

Agreed that a WebGL directory would be nice.

> > They are exactly the same. > > Or is there any concern for compilation time/cyclic import? cc @jdm Yeah, we shouldn't make net_traits depend on script_bindings, since that...

I've filed #36397 about that panic, because I'm worried that it has deep implications about our callback code and whether we enter realms appropriately.

``` ▶ CRASH [expected TIMEOUT] /IndexedDB/idbobjectstore-cross-realm-methods.html │ │ called `Result::unwrap()` on an `Err` value: Custom("invalid value: integer `16`, expected variant index 0