[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: multiple KDE segfault issues

On 17/11/11 06:36, Bob Tracy wrote:
> Just to eliminate KMS as a problem source and be able to perform a bit
> more KDE4 testing, I used FreeNX this morning to login remotely.  The
> plasma desktop looks *good* -- pretty much as I've gotten used to on my
> Kubuntu machines.  So far, everything is rock-solid stable.  I've
> resized the default desktop folder window, started a Konsole session and
> resized *that* window, and now have Iceweasel running as well.  I'm
> satisfied that KDE4 works and is stable on Alpha as long as the
> "qt4-x11" source package is built without optimization ("-O0" instead of
> the default "-O2").

Though -O0 is a huge performance hit.  Every variable access must be
read and written from memory and cannot be held in a CPU register.  That
cripples performance on a register based CPU such as the Alpha.

Would be interesting to try -O1 as that enables straight-forward
optimisations (such as allowing register variables) which will provide
quite a noticeable improvement in speed, without exercising the more
sophisticated optimisations of -O2.

Which version of qt4-x11 were you building? Unstable or experimental?

> Question for the community: how shall I report this upstream?  I'm
> tempted to work this on two fronts: report the workaround to the
> "qt4-x11" source package maintainer, and report the bug to the compiler
> maintainers. 

Yeah, probably best.  Maybe hold off to we can try a compilation with
-01.  I should be able to attempt that soon now that my ES45s have
arrived.  (Phew -- the freight company found them.  They sat in a
warehouse probably no more than a dozen kilometres away from me for over
a week, which is a longer time than they spent on a ship crossing the
2000 or so km from Australia to NZ.)

If it works with -O1 then it narrows down the search for the faulty

> The problem with the latter is, they're going to want a
> small test case that illustrates the bug:

Yes.  Maybe run a version of iceweasel with debugging symbols under gdb
until it segfaults.  Getting a backtrace is the first step in debugging.

I did see a very informative bug report for the segfault of iceweasel in
Lenny earlier this year and meant to go back to it and try the same
techniques as the bug reporter had illuminated, on a newer version of
iceweasel and the toolchain.  It's time to do that again but I now can't
find the bleeding bug report!


Reply to: