Created attachment 120902 [details]
gnumeric file causing segfault

Attaching the gnumeric file as per previous comment.

+ Trace 208411

I'll have a look.

I understand what happens, but it seems there are several issues around there, so it might take some time to fix.

Created attachment 120906 [details] [review]
should fix the issue

It appears that there are other issues concerning contour plots in trunk :-(
Anyway this bug seems fixed with this patch.

Comment on attachment 120906 [details] [review]
should fix the issue

Interacting directly with the klass is more than a bit nasty.   The followup safety checks seem prudent, but are also worrisome.  If they trigger is the resulting xls valid ?

I'll add a function to avoid the direct access to the klass. The resulting xls needs some testing. If it was not valid, we might just drop the plot (which would be easier anyway).

The resulting xls crashes gnumeric, but not calc.

Import crash filed separately as bug 557077.

Created attachment 121722 [details] [review]
new patch

This one does not access the klass. Resulting xls files can be opened in excel with no apparent issue, and in gnumeric, with some commited fixes to goffice.

Do we really need to call G_OBJECT_TYPE_NAME?  Is there a GO_IS_A_ macro
that could be used instead?

There is a macro, but it is defined in the plot_surface plugin and can't be used elsewhere unless we install the plugin headers.

Patch commited (both branches).