Re: [PATCH] Further extend "maint info sections" cmd with ALLOBJ

[Michael Snyder <msnyder@redhat.com>]

Pierre Muller wrote:
> 
> At 11:53 21/12/2001 , Pierre Muller a écrit:
> >At 23:30 20/12/2001 , Michael Snyder a écrit:
> >
> > >As a further extension, the "maint info sections" command
> > >will now accept an argument "ALLOBJ" to iterate over all
> > >known object files (which includes shared libraries.
> > >You can now do (for instance):
> > >
> > >         (gdb) maint info sect .bss ALLOBJ
> > >
> > >to see info on the .bss sections of all loaded object files.
> >
> >Great job!
> >
> >   Just a little remark, I would have expected that I get the same output
> >for the main executable in
> >
> >"maint info sect"
> >and
> >"maint info sect ALLOBJ"
> >but when I tried it out on a freshly compiled GDB,
> >the seciond command didn't should
> >the .stab and .stabstr sections that were shown for the first command.
> >
> >This is intentional?
> 
> The problem might be quite general :
> 
> maint info sect NEVER_LOAD
> gives the same output than
> maint info sect HAS_CONTENT
> 
> Debugging a little more lead me to find out that
> the reason of that bug is simply that
> NEVER_LOAD contains LOAD,
> ans strstr function returns thus true for strstr(string, "LOAD")
> if string is "NEVER_LOAD"

Oops!   ;-(

> I don't know the best way to solve this
> especially as someone could ask for both
> LOAD and NEVER_LOAD flags at the same time.
> (maybe adding a space a start of args and looking for ' LOAD' with a leading space).

Wouldn't work if "LOAD" was the only argument.

> Anyhow the current  code also accepts thingss like NOCODE for CODE section
> which is quite strange...)

Maybe I'll have to check for these specific cases.
I should have known that parsing wasn't this easy.  

> To come back to the problem about .stab section,
> the problem is related to the fact that
>    ALL_OBJFILE_OSECTIONS
> is defined as
> #define ALL_OBJFILE_OSECTIONS(objfile, osect)   \
>    for (osect = objfile->sections; osect < objfile->sections_end; osect++)
> 
> whereas the case without ALLOBJ is handled by a call to
> bfd_map_over_sections () function which does iterate in a different way:
>    for (sect = abfd->sections; sect != NULL; i++, sect = sect->next)
> 
> Wouldn't it be better to use the same bfd_map_over_sections ()
> function in the ALLOBJ case?

The problem is that ALLOBJ includes shared libraries, and
I wanted to print the relocated addresses for those, not the
raw addresses that are in the BFD section table.  I did not
see a way to get back from the bfd section to the objfile 
section, so I felt I had to start from the objfile section
in the first place.