[0/10] RFC: remove obj_section field

[0/10] RFC: remove obj_section field

[Tom Tromey]

This series removes general_symbol_info::obj_section.

The immediate benefit of doing this is that it shrinks all symbols by a
whole pointer.  This is a reasonably good memory savings.

This series also has the nice effect of regularizing section indices:
now the SYMBOL_SECTION is the same (modulo minor weirdness, see patch
#6) as the BFD section index.


This series also supports my long-term goal of "objfile splitting".  At
least patches 2, 3, and 4 go in this direction, by removing one
symbol->objfile backlink, and by introducing "bound minimal symbols".
I'll describe this more in those patches.


I've built and regtested this series on x86-64 Fedora 16, on PPC Fedora
18 (gcc110 in the compile farm), and on HP-UX.  I also built and
regtested it on x86-64 Fedora 16 using -gstabs+.

However, it could certainly use more testing and inspection.  The
individual patch messages will point out some of the questionable bits.

Joel, I will push a new branch for you to test, if you don't mind.  I'll
send email when it is ready.

Tom

Re: [0/10] RFC: remove obj_section field

[Joel Brobecker]

> Joel, I will push a new branch for you to test, if you don't mind.  I'll
> send email when it is ready.

Of course. I am really sorry I still haven't had the time to test it.
For the record, I agree to this being checked in HEAD prior to my
testing, and I will deal with the consequences after something ever
went wrong.

-- 
Joel

Re: [0/10] RFC: remove obj_section field

[Tom Tromey]

>>>>> "Joel" == Joel Brobecker <brobecker@adacore.com> writes:

>> Joel, I will push a new branch for you to test, if you don't mind.  I'll
>> send email when it is ready.

Joel> Of course. I am really sorry I still haven't had the time to test it.
Joel> For the record, I agree to this being checked in HEAD prior to my
Joel> testing, and I will deal with the consequences after something ever
Joel> went wrong.

I don't think it is that much of a rush to go to that extreme.
In practice (to my surprise) it has been a pretty easy branch to rebase.
And I'm sure we'll find bugs no matter when it goes in, I am fine with
taking the more patient approach.
I'll push the branch tomorrow.  Take your time testing it.

Tom

Re: [0/10] RFC: remove obj_section field

[Tom Tromey]

Tom> I'll push the branch tomorrow.  Take your time testing it.

I pushed it today.  It is still "archer-tromey-remove-obj_section" -- I
rebased it and did a force push.

Be warned that we're going to do a mass branch renaming in archer.git
soon.  After the renaming this will be "tromey/remove-obj_section".
If you can't find it using one name, look for the other name :)

Tom

Re: [0/10] RFC: remove obj_section field

[Joel Brobecker]

Hi Tom,

> Joel, I will push a new branch for you to test, if you don't mind.  I'll
> send email when it is ready.

I finally took the time to test your patch, and I am happy to say that
testing did not reveal any regression :).

-- 
Joel

Re: [0/10] RFC: remove obj_section field

[Pedro Alves]

Just wanted to say, great work, and that I'm all for this.

-- 
Pedro Alves

Re: [0/10] RFC: remove obj_section field

[Tom Tromey]

>>>>> "Tom" == Tom Tromey <tromey@redhat.com> writes:

Tom> This series removes general_symbol_info::obj_section.

I'm going to check this in now.

Please let me know if it causes problems.
I'll debug them as soon as I'm able.

This may conceivably cause build failures on some hosts.
I am happy to write patches for this as well, just send the errors in
email.

I rebased the series on today's trunk, re-built each patch, and
regtested the result.  So I think things should be reasonably ok.
In one case a patch needed a couple of changes after rebasing; I will
resent that shortly.

Tom

Re: [0/10] RFC: remove obj_section field

[Sandra Loosemore]

On 04/08/2013 01:44 PM, Tom Tromey wrote:
>>>>>> "Tom" == Tom Tromey <tromey@redhat.com> writes:
>
> Tom> This series removes general_symbol_info::obj_section.
>
> I'm going to check this in now.
>
> Please let me know if it causes problems.
> I'll debug them as soon as I'm able.
>
> This may conceivably cause build failures on some hosts.
> I am happy to write patches for this as well, just send the errors in
> email.

I'm seeing a build error:

.../gdb/symmisc.c: In function 'dump_msymbols':
.../gdb/symmisc.c:271:10: error: format '%ld' expects argument of type 
'long int', but argument 3 has type 'int' [-Werror=format]

-Sandra

Re: [0/10] RFC: remove obj_section field

[Tom Tromey]

>>>>> "Sandra" == Sandra Loosemore <sandra@codesourcery.com> writes:

Sandra> I'm seeing a build error:
Sandra> .../gdb/symmisc.c: In function 'dump_msymbols':
Sandra> .../gdb/symmisc.c:271:10: error: format '%ld' expects argument of type
Sandra> long int', but argument 3 has type 'int' [-Werror=format]

Oops, sorry about that.  Please try the appended.

Tom

--- symmisc.c.~1.95.~	2013-04-08 14:18:45.764743872 -0600
+++ symmisc.c	2013-04-08 19:13:07.605692663 -0600
@@ -268,7 +268,7 @@
 						section->the_bfd_section));
 	  else
 	    fprintf_filtered (outfile, " spurious section %ld",
-			      section - objfile->sections);
+			      (long) (section - objfile->sections));
 	}
       if (SYMBOL_DEMANGLED_NAME (msymbol) != NULL)
 	{

Re: [0/10] RFC: remove obj_section field

[Sandra Loosemore]

On 04/08/2013 07:15 PM, Tom Tromey wrote:
>>>>>> "Sandra" == Sandra Loosemore <sandra@codesourcery.com> writes:
>
> Sandra> I'm seeing a build error:
> Sandra> .../gdb/symmisc.c: In function 'dump_msymbols':
> Sandra> .../gdb/symmisc.c:271:10: error: format '%ld' expects argument of type
> Sandra> long int', but argument 3 has type 'int' [-Werror=format]
>
> Oops, sorry about that.  Please try the appended.
>
> Tom
>
> --- symmisc.c.~1.95.~	2013-04-08 14:18:45.764743872 -0600
> +++ symmisc.c	2013-04-08 19:13:07.605692663 -0600
> @@ -268,7 +268,7 @@
>   						section->the_bfd_section));
>   	  else
>   	    fprintf_filtered (outfile, " spurious section %ld",
> -			      section - objfile->sections);
> +			      (long) (section - objfile->sections));
>   	}
>         if (SYMBOL_DEMANGLED_NAME (msymbol) != NULL)
>   	{
>

This builds, at least.  :-)  Thanks.

-Sandra

Re: [0/10] RFC: remove obj_section field

[Tom Tromey]

Sandra> This builds, at least.  :-)  Thanks.

No problem.  I'm checking it in with this ChangeLog entry.

2013-04-08  Tom Tromey  <tromey@redhat.com>

	* symmisc.c (dump_msymbols): Cast fprintf_filtered argument to
	long.

Tom