On 30/09/2011, at 7:47 AM, Bob Tracy wrote:
On Thu, Sep 29, 2011 at 01:04:24PM -0400, Matt Turner wrote:
I presume there are patches to make it all work? Posting those is a
good place to start.
The original patches of Bob's were posted here, picked up by Magnus
who made some mods for Gentoo, which I sent back to Bob, and this is
Bob's second go at them.
other issues, the main impediment is Alpha no longer being a supported
architecture by the upstream code provider (Mozilla):
Presumably that is faced by a number of architectures.
one of the main
patches needed to get the Mozilla code to build on Alpha is
a downgrade of two source files to versions existing in firefox
We need someone who better understands the code to comment on that.
The Mozilla code "tortures" toolchains in ways that are hard to
reproduce using simple example code (as would be needed for useful bug
reports). We still have "-Wl,--no-relax" required as part of
That's a long-standing binutils problem. There are a growing number
of packages that are succumbing to that problem, including gcc-
snapshot --- maybe RTH might look at it now that Uros (the gcc Alpha
maintainer) has also reported it.
The optimizer on Alpha is broken, or at least behaves differently than
it evidently does on other 64-bit systems to which the Mozilla
maintainers have access: builds with "-O2" (default optimization)
typically segfault on invocation.
Which compiler version? The default gcc-4.4? Maybe gcc-4.5 or
gcc-4.6 would be better? There is a bug report of bad optimisation in
gcc-4.4 (#566195) which is, IIRC, fixed in gcc-4.5. (I've never got
around to adding to that bug report with extra info...)
I seem to recall an excellent bug report against iceweasel/xulrunner
in Lenny about the crash on invocation but when I looked a couple of
weeks ago to start following it up I could not find it. I meant to
repeat the tests reported in the bug report with the new wheezy
Other things on the "to do" list would include implementing the js JIT
code on Alpha:
I suspect that is a major exercise.
Still have to specify "--disable-ipc" as well, which isn't
necessary on supported architectures: that would need to be
and pushed upstream as well.
I think that would be relatively straight-forward to implement. Some
of the kernel code probably can be used as a model.
Matt: would you be interested in doing this? I suspect it is either
you or me who will have to do this and I am currently swamped with
being an Alpha buildd admin and fixing all the problems that result
from the three month's lost time building when we got demoted to
Again, I need to get in touch with Mike Hommey et. al. before this
can go much farther. I've added Mike to the "Cc:" list.
Yes, maybe Mike can take a look at your "fixes" and comment thereupon.
Bob, are you able to make available a source package with your
changes? That would be a lot less easier on your bandwidth to download.