Comment on attachment 444486 [details] [diff] [review] patch We should unify this with the nanojit detection code.

Question on irc about why auxv isn't working still stands -- this code has changed significnatly on m-c, when we pulled in a new pixman, and this won't apply; the new code still uses auxv, so if we can go that route, we should. The only issue was figuring out whether we could read /proc/self/auxv -- it seemed like we couldn't, but then maybe we could. It's entirely possible that we're just missing some permission in the manifest that would allow this (access to filesystems or something like that, basically access outside of our own /app/data directory).

WRITE_EXTERNAL_STORAGE would be one thing to test (and something we'll likely want to set anyway).

/proc/self/auxv is only readable by root, which /proc/cpuinfo is readable by all. I don't think this is a problem with the jvm permissions, but rather a user permissions problem. We can try to read auxv first and then fallback to reading cpuinfo, but we're going to be in the fallback case everytime as far as I can tell.

Also, Vlad mentioned that the 1024 char buffer might not be enough if we have multiple cores. The the cpu info from the nexus one takes 264 chars, and I would assume we'd get all the info we need off of the first core in the list anyway.

Here's a patch against trunk. Fwiw, we have the updated pixman on mozilla-droid.

Comment on attachment 445053 [details] [diff] [review] patch against trunk bleh, I guess it's fine.. just make sure you put a .patch file for pixman inside the cairo dir or this'll get blown away in the future. Also poke jeff to get this into upstream.