[PATCH v2 41/42] Make mapped_debug_names independent of objfile

[Simon Marchi]

On 2020-05-22 5:01 p.m., Tom Tromey wrote:
>>>>>> "Simon" == Simon Marchi via Gdb-patches <gdb-patches@sourceware.org> writes:
> 
> Simon> Ultimately, the objfile only seems to be needed because we might need to
> Simon> read a string from the string section.  For that, we might need to read
> Simon> in the section, and if it's a relocatable section, the objfile is needed
> Simon> in order to do the relocation.  This pattern happens often (that we to
> Simon> pass an objfile only because a section might be read).  I think it's a
> Simon> bit ugly, but I don't have a good alternative right now.
> 
> A long time ago, I changed the DWARF reader to read all the sections
> lazily.  Off and on I've wondered whether that was actually a mistake.
> For one thing, it requires passing stuff like this around.  Also, it can
> slow down hot paths, since they need to call the method to make sure the
> section is mapped.  There's also at least one bug in bugzilla related to
> this (if you unlink a .so, gdb can be unhappy).
> 
> For my threading project, I removed this lazy reading.  We should
> consider doing this sooner, perhaps.  The idea here is just that, if gdb
> is going to read the debuginfo anyway, then loading at least the
> sections needed to read .debug_info makes sense.

I think it would be interesting indeed.  Can we try to revive it after the
current patchset is merged?

Simon