Re: [MonetDB-users] python issues

On 16-11-2007 11:36:53 -0700, m h wrote:

On Nov 16, 2007 11:25 AM, Fabian Groffen wrote:

...

On 16-11-2007 10:33:29 -0700, m h wrote:

...

...

with the correct PYTHONPATH, the released ("-0") version also works for me (on FC6; both with Python 2.4.4 & 2.5)

Arghhh! Hmmmm, well I'm definitely crashing... Since this looks like a double free, perhaps this is a swig issue. (Ie, it's happening in the C not the python). What version of swig are you using? I'm on 1.3.31 (gentoo)....

Niels has had some fun on our Gentoo box and said he fixed it

http://monetdb.cvs.sourceforge.net/monetdb/clients/src/mapilib/Mapi.mx?view=log&pathrev=Clients_1-20#rev1.24.2.2

Wow, quick turnaround! Thanks!

So can Nels please elaborate on the issue? Why is this a Gentoo issue? Did he replicate the crash? Shouldn't other linuxes crash if there's a double free (that's what I grokked the error to be)?

I'm not sure if Fedora has glibc installed such that it makes double free detected messages.

I'm assumming no one else on the list is using the ebuild, hence no one else crashed?

Because our releases are always behind from the day they are released, most people here don't use any "packaged" versions, no ;)

...

Unfortunately that is in code which you cannot easily patch and recompile.

Well if I get motivated maybe I'll patch the file through the ebuild.... :)

That's the point. The patch is made to a .mx file, which is expanded into a .c and .h file. The ebuild only uses the .c and .h files, such that you don't need the "buildtools", which include a tool called "mx" which does extraction of .mx files. The only patch you could attempt, is to apply the patch to the corresponding .c file and hope patch will be able to match the hunks on the .c file code.

...

The easiest thing you can do now, is to wait an hour or 12 from now and use the monetdb-install.sh script with --nightly=stable --enable-sql to install a MonetDB/SQL from nightly source snapshots.

Doesn't the install script have a cvs pull option?

That too, but only for the cvs HEAD, as the script cannot know what the current stable branch is called.