This is the mail archive of the
mailing list for the GDB project.
Re: [patch] [3/5] Types reference counting [make_function_type-objfile]
- From: Jan Kratochvil <jan dot kratochvil at redhat dot com>
- To: Ulrich Weigand <uweigand at de dot ibm dot com>
- Cc: Tom Tromey <tromey at redhat dot com>, gdb-patches at sourceware dot org
- Date: Fri, 26 Jun 2009 18:39:27 +0200
- Subject: Re: [patch] [3/5] Types reference counting [make_function_type-objfile]
- References: <20090501144440.GA3213@host0.dyn.jankratochvil.net> <200906261323.n5QDN0Xi000973@d12av02.megacenter.de.ibm.com>
On Fri, 26 Jun 2009 15:23:00 +0200, Ulrich Weigand wrote:
> - I think the symbol readers should always return per-objfile types. This
> just makes everything simpler, in particular allocation of derived types
> (as you notice). I think it would also good to be able to guarantee that
> TYPE_OBJFILE of every symbol type is non-NULL ...
This IMO means "any two types somehow interconnected (by TYPE_TARGET_TYPE,
TYPE_INDEX_TYPE etc.) should have the same TYPE_OBJFILE".
> - The Java generated types should *not* go onto the "fake" dynamics objfile
> in the first place. That objfile is somewhat bogus in that it isn't
> associated with an architecture (thus breaking my per-type architecture
> effort), and it also don't help with type lifetime issues as the fake
> objfile never goes away. I think those types should be allocated with
> a NULL objfile (in the future are per-gdbarch types) instead.
I see now dynamics_objfile should be replaced by NULL TYPE_OBJFILE and being
a discardable type to be processed by the types garbage collector (which is
still not checked-in).
> Also, I think the implementation is broken in the "type smashing" case:
> if there is an incoming type allocated in objfile A, but the argument to
> make_function_type specifies objfile B,
I do not think it can happen. So far always type A referencing -> type
B either had the same TYPE_OBJFILE or TYPE_OBJFILE(B) was NULL. Using such
assertion-checks without facing problems.
As there cannot be a DWARF reference across objfiles (there can be one just
across CUs) cross-file type references use TYPE_IS_OPAQUE and they get
resolved each time on dereferencing such reference by calling check_typedef.
> Therefore, I'd prefer to fix the two problems mentioned above, and then
> revert your patch.
If the current rule:
# So far always type A referencing -> type B had either the same TYPE_OBJFILE
# or TYPE_OBJFILE(B) was NULL.
will be changed to:
# Type A referencing -> type B had the same TYPE_OBJFILE.
I only welcome it and sure I agree with reverting my patch afterwards.