Firefox 3.6 spidermonkey integration
Steven Parkes
It looks pretty simple at this point. There are some subtle issues with the split object that I'm still tracking down/checking but those probably don't concern anybody but me ... at least not yet.
There are a few tests that bork. Round-tripping through the parser is now removing some unnecessary parens. My inclination (and what I've done) is just to make either case acceptable for the tests. This is what I did in a case or two in the last integration. I presume this is on purpose from the SM folks and that we have to live with it ... but if someone else really cares ...
I've only look at the x64 Linux build so far. In the past that's been a bit of an issue and I'll be looking at x64 OS X.
I was even thinking of looking at cygwin ...
Steven Parkes
On Jan 26, 2010, at Jan 26,3:14 PM , Steven Parkes wrote:
> I've only look at the x64 Linux build so far. In the past that's been a bit of an issue and I'll be looking at x64 OS X.
Well, poop.
There appears to be a regression between 1.9.1 (the FF3.5 build) and 1.9.2 (the FF3.6 build). 1.9.2 won't build on OS X 10.6 x64. It generates bad assembly. The FF hg head does build ...
So we could
stick with 1.9.1
move to a snapshop of head
try to figure out what's messed in 1.9.2 and backport
I only just realized that the FF3.6 builds are all 32 bit ... 64 was mentioned for 3.5 in the past but seems to keep getting pushed out. Not that 64 bit support in all of FF keeps spidermonkey from supporting 64 bit, but if we don't have FF releases, we don't really have releases at all, I don't think.
Steven Parkes
On Feb 3, 2010, at Feb 3,7:28 PM , Steven Parkes wrote:
> There appears to be a regression between 1.9.1 (the FF3.5 build) and 1.9.2 (the FF3.6 build). 1.9.2 won't build on OS X 10.6 x64. It generates bad assembly. The FF hg head does build ...
Well, that was interesting. The regression was the old problem with detecting the wrong arch on OS X. It was building for 32 bits within a 64 bit build.
However, one thing this made me realize is that we haven't had JITing turned on, at least not on 64 bit builds (which is all I've been using lately.) It's turned off for 64 bit builds in the FF 3.5 and 3.6 releases.
So I created a jit branch with the current head of mozilla-central, which is kinda of head for FF, as I understand it.
It works. I was able to turn on debugging to convince myself it really is JITing.
But I don't know what this means for our next release. I'm a little reluctant to use something based on a snapshot of head. But I can't figure out what Moz is going to do with the next version of FF. I saw some indication they were going to skip 3.7 and go to 4, but then 3.7 is mentioned in recent commits.