This is the mail archive of the
gdb-patches@sourceware.org
mailing list for the GDB project.
Re: Removing TYPE_VPTR_FIELDNO uses (was: Re: [patch ping] SetTYPE_VPTR_BASETYPE/TYPE_VPTR_FIELDNO of XL C++ virtual class)
- From: Wu Zhou <woodzltc at cn dot ibm dot com>
- To: Daniel Jacobowitz <drow at false dot org>
- Cc: gdb-patches at sources dot redhat dot com
- Date: Sun, 9 Oct 2005 12:02:16 +0800 (CST)
- Subject: Re: Removing TYPE_VPTR_FIELDNO uses (was: Re: [patch ping] SetTYPE_VPTR_BASETYPE/TYPE_VPTR_FIELDNO of XL C++ virtual class)
- References: <1127969598.433b733eedab9@imap.linux.ibm.com><20051002222103.GA32728@nevyn.them.org>
Hi Daniel,
Thanks for your comments. Some more questions about these code.
> > - gdbtypes.c/gdbtype.h: to initialize VPTR_FIELDNO (in alloc_type and
> > create_array_type), fill VPTRs (in fill_in_vptr_fieldno), and dump VPTRs
> > (in recursive_dump_type). Maybe some change to the type dumping is
> > needed.
>
> Not if we leave them for older ABIs and stabs.
Maybe we need to add some code to dump VPTRs for gnu-v3 ABI after removing
TYPE_VPTR_FIELDNO?
> > - eval.c (evaluate_subexp_standard): TYPE_VPTR_BASETYPE is used to iterate
> > the baseclasses to find the real address of the virtual function.
>
> This code needs to be (A) read and thought about, so that we can figure
> out what it used to do, and (B) replaced with something less broken.
> It hasn't worked in forever. Take a look at what METHOD_PTR_IS_VIRTUAL
> expands to!
It seems that the definition for METHOD_PTR_IS_VIRTUAL is at least error
for 64-bit arch. Seen from the changelogs I found it was introduced in
gdb since 1992. Will this still stands after more than ten years?
#define METHOD_PTR_IS_VIRTUAL(ADDR) ((ADDR) & 0x80000000)
Didn't all these different compilers reached an agreement on how to
predicate a pointer-to-method is virtual?
> > - gnu-v3-abi.c: VPTRs is used for rtti, virtual function and virtual base
> > class offset.
>
> In this file, I think we can skip all the rigamarole with debug
> information to find the vptr. It's very easy: it's the size of a
> pointer, and it's at offset 0. Always.
If so, we can use this to find the vptr.
> Unless of course there isn't one. We may need to figure out what the
> field at offset 0 is to see whether it's a vptr or a user variable. I
> haven't thought about that in a while; maybe we can assume that there
> is one by the time we get into this file.
Do you have any clues on how to determine whether this assumption stands?
Best Regards
- Wu Zhou