This is the mail archive of the gdb@sources.redhat.com mailing list for the GDB project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: add-symbol-file-from-memory patch breaks non-elf targets?


Roland McGrath <roland@redhat.com> writes:

It's unfortunate that this is done in a way that requires a tweak for every
specific target. The functionality will work, and thus ought to be
available, on any ELF target. The use of symbol_file_add_from_memory from
other gdb code to support Linux's vDSO is already needed on more than one
Linux target and it is not unlikely that other Linux targets will follow
suit in the future. It is crufty and makes maintenance error prone if we
have nothing better than copying and updating makefile boilerplate for
every specific Linux/CPU target. For example, your patch covers only x86
and powerpc, but already IA64 and x86-64 need this support (and AFAIK not
yet powerpc).


But "any ELF target" is not a well-defined thing in GDB.  GDB just
tries to cope with whatever it gets.  GDB lets BFD keep track of what
object file formats are appropriate for what targets.

The reason I thought this approach would be appropriate is that it
resembles the way we handle the different dynamic linker debugging
protocols: each .mt file selects solib-mumble.o manually.  As far as I
know, this hasn't been much of a maintenance hassle.

No, that's also a bug. As part of multi-arch, GDB should be able to include multiple shlib mechanisms.


Anyway, looks like I'm fixing this.

Andrew



Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]