[RFC] PR symtab/34741: expression lookup and overridden weak functions

[RFC] PR symtab/34741: expression lookup and overridden weak functions

[Longjun Luo via Gdb]

Hi,

I filed PR symtab/34741:
https://sourceware.org/bugzilla/show_bug.cgi?id=34741

With ordinary compiler output, GDB can call an overridden weak function
instead of the strong definition used by the program. In the reproducer,
"print action()" returns 22 while the program's call returns 11. This
happens in the weak definition's CU and, depending on CU search order,
in unrelated CUs too.

For ordinary global function lookup, my local fix uses the main ELF
executable's minimal symbol to select a full definition with the same
linkage name and address, taking both type and address from it. Lexical
and explicit file-qualified lookups are unchanged. Does this direction
look reasonable?

I have no FSF copyright assignment for GDB. The implementation is a few
hundred lines plus regression tests, so it is not a trivial change.
Would a DCO sign-off be acceptable, or should I arrange an assignment?

Thanks,
Longjun Luo

Re: [RFC] PR symtab/34741: expression lookup and overridden weak functions

[Tom Tromey]

>>>>> Longjun Luo via Gdb <gdb@sourceware.org> writes:

I didn't see a reply to this.

> For ordinary global function lookup, my local fix uses the main ELF
> executable's minimal symbol to select a full definition with the same
> linkage name and address, taking both type and address from it. Lexical
> and explicit file-qualified lookups are unchanged. Does this direction
> look reasonable?

I don't really know this area well, so I can't comment on this; but here:

> I have no FSF copyright assignment for GDB. The implementation is a few
> hundred lines plus regression tests, so it is not a trivial change.
> Would a DCO sign-off be acceptable, or should I arrange an assignment?

I think you should get an assignment.

thanks,
Tom