Re: Issue 1809 in lilypond: make test segfaults

[lilypond]

Comment #1 on issue 1809 by address@hidden: make test segfaults
http://code.google.com/p/lilypond/issues/detail?id=1809

This is a somewhat frustrating state of things.  I reconfigured with
./configure CXXFLAGS="-fkeep-inline-functions -fno-jump-threading"

in order to get a more useful debugging environment
(keep-inline-functions keeps a lot of things like array constructors
in code around so that the debugger can access arrays and stuff, and
no-jump-threading makes sure that stack backtraces are actually
pointing to the right function instead of something that the compiler
considered fit for code sharing).  Apart from discovering that
Simple_spacer was missing implementations of unused functions and
members (it worked by chance since the compiler did not bother to
complain about missing things that were unused), this did not help
much:

The tests just completed fine after that.  Configuring and compiling
without those options (which have no effect that should be relevant to
the problem) again lead back to the situation with the coredumps.

I can't always reproduce them, and recently they refused occuring
altogether when the program is run in the debugger (post mortem
debugging on the core file is not all that helpful with all the C++
class folderol when I don't have the inline accessor functions
available).

The core dumps seem partly related to corrupted garbage collection
data.  They tend to occur frequently in connection with corruption of
Grob_info data, and when translator-dispatch-list is cycling through
the function pointers and engravers.

I don't have a good idea how to further delve into this.  Open for
suggestions as long as the crashes are somewhat reliably there.