This is the mail archive of the
gdb-patches@sourceware.org
mailing list for the GDB project.
Re: [PATCH 11/18] Add partition support.
>>>>> "Joel" == Joel Brobecker <brobecker@adacore.com> writes:
Joel> +/* We need to keep the symbols in each PD separate from the others.
Joel> + One of the reasons is that 2 symbols with the same name are allowed to
Joel> + coexist in 653 provided that these 2 symbols live in 2 separate PDs.
Joel> + Another important reason is that virtual memory regions can overlap
Joel> + between PDs, and one symbol address in a PD is not suitable in the
Joel> + context of another PD.
Joel> +
Joel> + GDB maintains a global list of objfiles where the symbols can be found.
Joel> + To achieve this separation, we simply maintain one such list per PD in
Joel> + this module. At each PD change, we simply save the objfile list for
Joel> + the current PD, and restore it for the new PD. The net effect is that
Joel> + GDB will only recognize symbols in the current PD, and will complain if
Joel> + a user tries to do any operation that needs symbols from other PDs.
Joel> +
Joel> + In practice, GDB maintains a few other global variables that also
Joel> + need to be saved/restored, but the principle remains valid.
Joel> +
Joel> + If would have been slicker to actually maintain one list of objfiles
Joel> + for all PDs and modify GDB to automatically switch to the right PD
Joel> + when performing an operation such as inserting a BP in some code that's
Joel> + in a different partition. This would probably not be such a great
Joel> + improvment in terms of usability, but this would require a major
Typo, "improvement". Thanks, Emacs!
More importantly, why can't a PD be modeled using program- and
address-spaces inside GDB?
Tom