Re: [PATCH,V4 2/2] CTF: multi-CU and archive support

[Weimin Pan <weimin.pan@oracle.com>]

On 5/10/2021 2:27 PM, Tom Tromey wrote:
>>>>>> ">" == Wei-min Pan <weimin.pan@oracle.com> writes:
>>> Thank you for taking the time. Please see below for some explanation
>>> about CTF and let me know if you need more info.
> Thank you.
>
>>> Where each archive has the CU as its name and ".ctf" as its parent's name
>>> (emitted by the linker.) In terms of psymtabs, there is only one parent
>>> which can have any number of archives.
> I think this will probably yield over-expansion in gdb.  What I mean is
> that if any symbol is found in CTF, all types and symbols will be
> expanded.  Normally this isn't what you want, though -- you only really
> need or want to expand the psymtab that has the type (or whatever) that
> was asked for.
>
> Normally what you want is for a single psymtab to correspond to a single
> source file.  In gdb the dependencies are used for #includes, which I
> guess comes from how stabs were originally implemented.

Yes, that's what happens with the CTF layout. While there is some
benefit of generating psymtabs, it's quite limited as you pointed out.
And I doubt that even getting rid of emitting psymtabs altogether
will not help much. Do you have any suggestions?

Thanks.

>
> Tom