Given bugs like bug 396052, we should at least be trying to launch the browser from a non-ASCII path. We could possibly do this on the perftest machines. It'd be nice to avoid future bustage like this.

Also it looks like we're broken on Win32 with a non-ascii username, so that'd be good to test too.

See also bug 396199.

Moving to Talos component.

Is this worth having a whole test devoted to it? Sounds more like a unit test then a talos test, is all.

My original summary was simply "should run basic startup tests from non-ASCII paths". It's kind of a PITA to write tests that have to start the browser specially right now, since we don't have a harness for that, but I don't really care if this were to be Talos or just some other new harness.

This is actually not very complicated. If you add "!!python/str" tag in front of the browser_path definition in the yml file and ensure that the yml file is encoded in utf-8, it works correctly. One other way is to loop over all the parsed yaml values and decode them to utf-8 encoded strings instead of unicode strings. Changing setEnvironmentVars to use str decoding is not a good option since the parameters passed can be instances of int or some other class.

Adding jhammel since this may be related with his efforts on PerfConfigurator.

this isn't needed for a performance test, lets resolve as a unittest.