On 5/13/14, 10:11 AM, James Graham wrote:
I think the problem that I have with this API is "the number of cores that exist" isn't obviously a good proxy for "the number of cores that are available". It I have N cores and am already using M cores for e.g. decompressing video, N-M is probably a much better estimate of the available resources than N.
One thing along those lines that came up in discussion of this proposal at Mozilla: this proposal seems to assume that the UA itself is using a very few cores. This is a valid-ish assumption for current UAs: as long as you have only one tab open, the UA will typically use only one core (ignoring things like parallel JIT compilation, parallel GC, etc; I did say "valid-ish"). But I suspect that for future UAs this will be an outright bad assumption. Certainly for Servo it is.
-Boris
