Bug#225146: segfault in crm114 when using classify
On Fri, 2 Jan 2004, Miller, Raul D wrote: > In my mind, the right way to deal with this problem is: > [3] When the author releases 1.0, release that as a new package under > a different name (perhaps "crm") which conflicts with the alpha crm114 > package. Also, at that time, update the alpha crm114 package's > description (and NEWS) files with a note that the package is alpha, > that it's been superceeded by the version 1.0 instance of crm, and that > the .css file formats are incompatible. Please don't do this. The package is called crm114, calling it crm will confuse many, and cause many who search for crm114 to not find it. If you really need to keep both versions of the package around, upload a package called crm114-alpha that Replaces crm114 (<= 2002-11-26-3) and Conflicts with crm114. This effectively renames the package, and allows those who need it to keep the old .css format. Put a message in the control file mentioning that this version is very old, and to use it only if you must. Then during debconf of the new crm114 if you notice you are upgrading from 2002-11-26-3 or less pop up a message (level critical) telling them that the format for .css has changed and to retrain by running: learnspam.crm < spamtext.txt and learnnonspam.crm < nonspamtext.txt Personally I think you should just forget about maintaining compatibility with the old format for .css - if you have spamtext.txt and nonspamtext.txt then you don't need it, and if you don't then just start over - the old version of crm114 is much less accurate anyway. -Ariel
Bug#225146: segfault in crm114 when using classify
Colin Watson wrote: > Looks like malloc arena corruption. Could you install valgrind and run > 'valgrind --num-callers=20 crm114 -t bin/incoming.crm That should show up the location of the corruption. I haven't gotten around to trying that version of crm with vagrind, yet. I have, however, been in contact with the developer -- he told me that the alpha version which we have as our release version has had two significant subsequent releases. Also, he has released two more versions specifically in the context of this segfault bug [and another bug which I reported directly to him]. (these are: crm114-20031229-1.0-SanityCheck.src.tar.gz and crm114-20040102-1.0-SanityCheck.src.tar.gz). The classify bug appears to be no longer present in the 20031229 version, but it's difficult to tell because the .css file format has changed [so I couldn't properly test that bug -- it is only triggered with specific .css files]. In my mind, the right way to deal with this problem is: [1] Document that this problem exists, and that it can be worked around by retraining the .css files with additional content. [2] Downgrade the severity of this bug to normal. [3] When the author releases 1.0, release that as a new package under a different name (perhaps "crm") which conflicts with the alpha crm114 package. Also, at that time, update the alpha crm114 package's description (and NEWS) files with a note that the package is alpha, that it's been superceeded by the version 1.0 instance of crm, and that the .css file formats are incompatible. I'll do this myself if no one else gets around to it first [but I'm waiting to get my bugs resolved before I put any time into this]. Thanks, -- Raul
Bug#225146: segfault in crm114 when using classify
On Fri, Dec 26, 2003 at 04:03:17PM -0500, Miller, Raul D wrote: > Package: crm114 > Version: 2002-11-26-2 > Severity: grave > > glibc6 version: 2.3.2.ds1-10 > > crm114 segfaults on me, when I use classify. Here's what it looks > like when I run strace crm114 -t bin/incoming.crm Here's what gdb tells me about that segfault when I compile from > source and run the unstripped binary: > > Program received signal SIGSEGV, Segmentation fault. > 0x4009b10b in mallopt () from /lib/libc.so.6 Looks like malloc arena corruption. Could you install valgrind and run 'valgrind --num-callers=20 crm114 -t bin/incoming.crm
