tags 235328 patch thanks Hi Kevin, On Sat, Feb 28, 2004 at 11:23:53PM -0600, Kevin Corry wrote: > Thanks for the patch. However, all of the dlist code has been removed from > EVMS in the CVS tree (replaced with much simpler list services), which is > where all of the bugs seem to be occurring. So the CVS tree (and thus the > 2.3.0 release) will hopefully be free of these problems. > If you have some time, we'd greatly appreciate if you'd be able to grab the > current CVS tree and test on your alpha system to see if you experience any > similar problems. You can find info about downloading the CVS tree at: > http://sourceforge.net/cvs/?group_id=25076 I've been able to reproduce the bug locally, and have thus been able to confirm both that the patch in my previous mail corrects the problem, and that the 2.3 CVS code does not contain the bug (as expected). This is basically as far as I've gone with testing 2.3 CVS, since alpha-style BSD disklabels don't seem to be supported yet, but building with -Wcast-align also leaves me confident that there have been no alpha-related regressions between 2.2 and 2.3. Regards, -- Steve Langasek postmodern programmer > On Saturday 28 February 2004 22:20, Steve Langasek wrote: > > On Sat, Feb 28, 2004 at 01:45:52PM -0800, Matt Zimmerman wrote: > > > On Sat, Feb 28, 2004 at 11:57:58AM -0600, Steve Langasek wrote: > > > > I know that evms 1.2 had problems such as this, where some unfortunate > > > > casts resulted in 64-bit values being written to the address of a > > > > 32-bit field. I'm not currently using evms on alpha, but I can > > > > probably dig up some patches to see whether they still apply to evms > > > > 2.2. > > > > > > > > This ought to be filed as a grave bug against the Debian package, FWIW. > > > > > > This is #235328 now, and if you have any insight or patches, by all means > > > send them to the BTS. > > > > Ok, I resurrected the retired box long enough to pull my code changes > > off of it, and sure enough, evms 2.2 has a 64-bit bug in the same place > > as the evms 1.2 codebase. This time I did a search through the code to > > look for the same error in other parts of the code; attached is a patch > > which addresses all instances that I found. Apologies for not having > > submitted this patch long ago. > > > > FWIW, I could not reproduce the segfault in question on my current > > system, probably because all my disks are using BSD disklabels, which > > EVMS does not appear to understand; so this patch *might* not fix the > > segfault in question -- but odds are good that it does. If not, it > > might be a good idea to build the whole system with -Wcast-align once, > > to see what else shows up. > > > > A bit of friendly 64-bit advise to the EVMS folks: if you have an > > explicit data type in your code, don't substitute an arbitrary integer > > type for it in your declarations; and if you *do*, at least don't use an > > explicit cast to suppress the compiler warnings when you try to use its > > address. ;) > > > > Cheers,
Attachment:
signature.asc
Description: Digital signature