Re: New ABI demangler support in binutils

Re: New ABI demangler support in binutils

[Nick Clifton]

Hi H.J.

: > The demangler support for the new API is available in the libiberty that
: > comes with GCC, but does not seem to be available in binutils. How does
: > one go about getting demangler support for the new ABI in binutils? How
: > is libiberty kept in sync between GCC, gdb, and binutils?
: > 
: 
: Someone has to sync it.

Correct, althpough there are (tentative) plans to merge the gcc abnd
binutils repositories at some point in the future.

: Nick, should I give it a try?

Please do.

Cheers
	Nick

Re: New ABI demangler support in binutils

[H . J . Lu]

On Thu, Jun 15, 2000 at 11:09:47AM -0700, Nick Clifton wrote:
> Hi H.J.
> 
> : > The demangler support for the new API is available in the libiberty that
> : > comes with GCC, but does not seem to be available in binutils. How does
> : > one go about getting demangler support for the new ABI in binutils? How
> : > is libiberty kept in sync between GCC, gdb, and binutils?
> : > 
> : 
> : Someone has to sync it.
> 
> Correct, althpough there are (tentative) plans to merge the gcc abnd
> binutils repositories at some point in the future.
> 
> : Nick, should I give it a try?
> 
> Please do.
> 

Should I sync the whole libiberty or just the demangler? I prefer
the whole libiberty since there are quite a few files for demangler
in libiberty and I don't know if they will work right without other
changes.


H.J.

Re: New ABI demangler support in binutils

[Nick Clifton]

Hi H.J.

: Should I sync the whole libiberty or just the demangler? I prefer
: the whole libiberty since there are quite a few files for demangler
: in libiberty and I don't know if they will work right without other
: changes.

How much work is involved and how likely is it that this will break
things ?

Ideally I would like to see the two versions of the library's sources
being identical, but I am worried that this mightbreak too many
things.  Also I am not sure if the gcc libiberty should be considered
the master copy and the binutils one just a shadow of it. or if there
are important features in both which need to be merged together.

If you have the time and the desire then I would suggest that you do
try to merge the whole of the two libiberties.  But please be careful!

Cheers
	Nick

Re: New ABI demangler support in binutils

[Ken Block]

Everything would probably be best, but the following is the minimum you can take
for the demangler:


libiberty/Makefile.in
libiberty/cplus-dem.c
libiberty/cp-demangle.c (new)
libiberty/dyn-string.c (new)
libiberty/sort.c (new)

include/demangle.h
include/sort.h (new)
include/dyn-string.h (new)



> On Thu, Jun 15, 2000 at 11:09:47AM -0700, Nick Clifton wrote:
> > Hi H.J.
> >
> > : > The demangler support for the new API is available in the libiberty that
> > : > comes with GCC, but does not seem to be available in binutils. How does
> > : > one go about getting demangler support for the new ABI in binutils? How
> > : > is libiberty kept in sync between GCC, gdb, and binutils?
> > : >
> > :
> > : Someone has to sync it.
> >
> > Correct, althpough there are (tentative) plans to merge the gcc abnd
> > binutils repositories at some point in the future.
> >
> > : Nick, should I give it a try?
> >
> > Please do.
> >
>
> Should I sync the whole libiberty or just the demangler? I prefer
> the whole libiberty since there are quite a few files for demangler
> in libiberty and I don't know if they will work right without other
> changes.
>
> H.J.

Re: New ABI demangler support in binutils

[Ian Lance Taylor]

   Date: Thu, 15 Jun 2000 12:15:55 -0700
   From: Nick Clifton <nickc@cygnus.com>

   : Should I sync the whole libiberty or just the demangler? I prefer
   : the whole libiberty since there are quite a few files for demangler
   : in libiberty and I don't know if they will work right without other
   : changes.

   How much work is involved and how likely is it that this will break
   things ?

   Ideally I would like to see the two versions of the library's sources
   being identical, but I am worried that this mightbreak too many
   things.  Also I am not sure if the gcc libiberty should be considered
   the master copy and the binutils one just a shadow of it. or if there
   are important features in both which need to be merged together.

   If you have the time and the desire then I would suggest that you do
   try to merge the whole of the two libiberties.  But please be careful!

The libraries should be identical.  I used to periodically simply
import the gcc libiberty into the binutils libiberty.  No changes
should be checked into the binutils libiberty without being first
checked into the gcc libiberty.

Ian

Re: New ABI demangler support in binutils

[Mark Mitchell]

>>>>> "Ken" == Ken Block <block@zk3.dec.com> writes:

    Ken> Everything would probably be best, but the following is the
    Ken> minimum you can take for the demangler:


    Ken> libiberty/Makefile.in libiberty/cplus-dem.c
    Ken> libiberty/cp-demangle.c (new) libiberty/dyn-string.c (new)
    Ken> libiberty/sort.c (new)

    Ken> include/demangle.h include/sort.h (new) include/dyn-string.h
    Ken> (new)

Actually, you don't need sort.c or sort.h, unless I'm mistaken.  Those
are specialized routines for sorting arrays of pointers, and I'm
pretty sure the demangler doesn't use them.

--
Mark Mitchell                   mark@codesourcery.com
CodeSourcery, LLC               http://www.codesourcery.com

Re: New ABI demangler support in binutils

[Ken Block]

If you don't take sort, you will need to modify makefile.in since it
references it.

Mark Mitchell wrote:

> >>>>> "Ken" == Ken Block <block@zk3.dec.com> writes:
>
>     Ken> Everything would probably be best, but the following is the
>     Ken> minimum you can take for the demangler:
>
>     Ken> libiberty/Makefile.in libiberty/cplus-dem.c
>     Ken> libiberty/cp-demangle.c (new) libiberty/dyn-string.c (new)
>     Ken> libiberty/sort.c (new)
>
>     Ken> include/demangle.h include/sort.h (new) include/dyn-string.h
>     Ken> (new)
>
> Actually, you don't need sort.c or sort.h, unless I'm mistaken.  Those
> are specialized routines for sorting arrays of pointers, and I'm
> pretty sure the demangler doesn't use them.
>
> --
> Mark Mitchell                   mark@codesourcery.com
> CodeSourcery, LLC               http://www.codesourcery.com

Re: New ABI demangler support in binutils

[Mark Mitchell]

>>>>> "Ken" == Ken Block <block@zk3.dec.com> writes:

    Ken> If you don't take sort, you will need to modify makefile.in
    Ken> since it references it.

True.

Nick, these changes should be very safe.  There are almost no
modifications to the old demangler; there's just a brand-new
additional source file.
 
--
Mark Mitchell                   mark@codesourcery.com
CodeSourcery, LLC               http://www.codesourcery.com

Re: New ABI demangler support in binutils

[Chris Faylor]

On Thu, Jun 15, 2000 at 12:15:55PM -0700, Nick Clifton wrote:
>Hi H.J.
>
>: Should I sync the whole libiberty or just the demangler? I prefer
>: the whole libiberty since there are quite a few files for demangler
>: in libiberty and I don't know if they will work right without other
>: changes.
>
>How much work is involved and how likely is it that this will break
>things ?
>
>Ideally I would like to see the two versions of the library's sources
>being identical, but I am worried that this mightbreak too many
>things.  Also I am not sure if the gcc libiberty should be considered
>the master copy and the binutils one just a shadow of it. or if there
>are important features in both which need to be merged together.
>
>If you have the time and the desire then I would suggest that you do
>try to merge the whole of the two libiberties.  But please be careful!

It probably goes without saying, but I'll ask it anyway -- when you copy
libiberty you also will have to move include files from the top level
include directory.  Should the include directory be moved over
completely as well or should only specific files be copied.

cgf

Re: New ABI demangler support in binutils

[H . J . Lu]

On Thu, Jun 15, 2000 at 12:23:15PM -0700, Ian Lance Taylor wrote:
> 
> The libraries should be identical.  I used to periodically simply
> import the gcc libiberty into the binutils libiberty.  No changes
> should be checked into the binutils libiberty without being first
> checked into the gcc libiberty.

I synced my libiberty with gcc. I built/checked it under Linux/ia32.
I am doing those on Linux/alpha, Solaris/ia32 and Solaris/Sparc now.
Assuming everything is ok, how should I sync it? Should I use
"cvs import" or "cvs add/ci"? I prefer "cvs add/ci".

Thanks.

H.J.

Re: New ABI demangler support in binutils

[Ian Lance Taylor]

   Date: Thu, 15 Jun 2000 13:31:17 -0700
   From: "H . J . Lu" <hjl@lucon.org>

   On Thu, Jun 15, 2000 at 12:23:15PM -0700, Ian Lance Taylor wrote:
   > 
   > The libraries should be identical.  I used to periodically simply
   > import the gcc libiberty into the binutils libiberty.  No changes
   > should be checked into the binutils libiberty without being first
   > checked into the gcc libiberty.

   I synced my libiberty with gcc. I built/checked it under Linux/ia32.
   I am doing those on Linux/alpha, Solaris/ia32 and Solaris/Sparc now.
   Assuming everything is ok, how should I sync it? Should I use
   "cvs import" or "cvs add/ci"? I prefer "cvs add/ci".

I can't remember which I did in the past.  Probably just `cvs add/ci'.
If I did `cvs import', you'll be able to see that in the `cvs log'
output.

Ian

Re: New ABI demangler support in binutils

[H . J . Lu]

On Thu, Jun 15, 2000 at 01:33:09PM -0700, Ian Lance Taylor wrote:
>    Date: Thu, 15 Jun 2000 13:31:17 -0700
>    From: "H . J . Lu" <hjl@lucon.org>
> 
>    On Thu, Jun 15, 2000 at 12:23:15PM -0700, Ian Lance Taylor wrote:
>    > 
>    > The libraries should be identical.  I used to periodically simply
>    > import the gcc libiberty into the binutils libiberty.  No changes
>    > should be checked into the binutils libiberty without being first
>    > checked into the gcc libiberty.
> 
>    I synced my libiberty with gcc. I built/checked it under Linux/ia32.
>    I am doing those on Linux/alpha, Solaris/ia32 and Solaris/Sparc now.
>    Assuming everything is ok, how should I sync it? Should I use
>    "cvs import" or "cvs add/ci"? I prefer "cvs add/ci".
> 
> I can't remember which I did in the past.  Probably just `cvs add/ci'.
> If I did `cvs import', you'll be able to see that in the `cvs log'
> output.

It looks like `cvs add/ci' used at least on filenames.h. I will
use it also.

Linux/alpha, Solaris/ia32 and Solaris/Sparc are all ok. I will sync
sourceware and verify it now. It may take an hour or so.


H.J.

Re: New ABI demangler support in binutils

[H . J . Lu]

On Thu, Jun 15, 2000 at 11:09:47AM -0700, Nick Clifton wrote:
> Hi H.J.
> 
> : > The demangler support for the new API is available in the libiberty that
> : > comes with GCC, but does not seem to be available in binutils. How does
> : > one go about getting demangler support for the new ABI in binutils? How
> : > is libiberty kept in sync between GCC, gdb, and binutils?
> : > 
> : 
> : Someone has to sync it.
> 
> Correct, althpough there are (tentative) plans to merge the gcc abnd
> binutils repositories at some point in the future.
> 
> : Nick, should I give it a try?
> 
> Please do.

Done. Verified on Linux/ia32. There are no regressions. Let me know if
there are any problems related to it.

Thanks.


H.J.

Re: New ABI demangler support in binutils

[Chris Faylor]

On Thu, Jun 15, 2000 at 02:05:31PM -0700, H . J . Lu wrote:
>On Thu, Jun 15, 2000 at 11:09:47AM -0700, Nick Clifton wrote:
>> Hi H.J.
>> 
>> : > The demangler support for the new API is available in the libiberty that
>> : > comes with GCC, but does not seem to be available in binutils. How does
>> : > one go about getting demangler support for the new ABI in binutils? How
>> : > is libiberty kept in sync between GCC, gdb, and binutils?
>> : > 
>> : 
>> : Someone has to sync it.
>> 
>> Correct, althpough there are (tentative) plans to merge the gcc abnd
>> binutils repositories at some point in the future.
>> 
>> : Nick, should I give it a try?
>> 
>> Please do.
>
>Done. Verified on Linux/ia32. There are no regressions. Let me know if
>there are any problems related to it.

I should have mentioned before that the last time someone synced the
two, it broke various things in Cygwin.  I think that DJ has been trying
to get a patch approved in gcc-land to fix this.  It is approved, I
believe, but it doesn't seem to have been checked in.

DJ, is this correct?  Is libiberty broken again for cygwin?

cgf

Re: New ABI demangler support in binutils

[DJ Delorie]

> DJ, is this correct?  Is libiberty broken again for cygwin?

The checkin did not include the make/configure files, just the ones
that were needed for the particular feature they needed.

My patch is 1/2 approved.  The other half affects native cygwin
builds, so I'm still waiting.

Re: New ABI demangler support in binutils

[David O'Brien]

On Thu, Jun 15, 2000 at 01:31:17PM -0700, H . J . Lu wrote:
> Assuming everything is ok, how should I sync it? Should I use
> "cvs import" or "cvs add/ci"? I prefer "cvs add/ci".

I would prefer a CVS repo copy.  This way commit logs have some meaning,
and changes may be easily tracked.
 
-- 
-- David    (obrien@NUXI.com)

Re: New ABI demangler support in binutils

[Andrew Cagney]

"H . J . Lu" wrote:

> 
> Should I sync the whole libiberty or just the demangler? I prefer
> the whole libiberty since there are quite a few files for demangler
> in libiberty and I don't know if they will work right without other
> changes.

FYI, GDB couldn't care.  Just remember to let us know when it has been
done.

	Andrew

Re: New ABI demangler support in binutils

[Andrew Cagney]

David O'Brien wrote:
> 
> On Thu, Jun 15, 2000 at 01:31:17PM -0700, H . J . Lu wrote:
> > Assuming everything is ok, how should I sync it? Should I use
> > "cvs import" or "cvs add/ci"? I prefer "cvs add/ci".
> 
> I would prefer a CVS repo copy.  This way commit logs have some meaning,
> and changes may be easily tracked.

Unfortunatly that would throw away all the CVS tags that binutils, gdb
and the like have been adding to their local repository.

	Andrew

Re: New ABI demangler support in binutils

[H . J . Lu]

On Tue, Jun 20, 2000 at 01:00:02PM +1000, Andrew Cagney wrote:
> "H . J . Lu" wrote:
> 
> > 
> > Should I sync the whole libiberty or just the demangler? I prefer
> > the whole libiberty since there are quite a few files for demangler
> > in libiberty and I don't know if they will work right without other
> > changes.
> 
> FYI, GDB couldn't care.  Just remember to let us know when it has been
> done.

It is done now.


H.J.

Re: New ABI demangler support in binutils

[Nick Clifton]

Hi Chris,

: >Really ?  Do you mean all of the files in include/ or just
: >include/libiberty.h ?
: 
: Here's a crudely derived list:
: 
: ansidecl.h
: demangle.h
: dyn-string.h
: floatformat.h
: getopt.h
: hashtab.h
: libiberty.h
: objalloc.h
: obstack.h
: partition.h
: sort.h
: splay-tree.h
: 
: These include files are referenced in the gcc 'libiberty' and also
: present in 'include'.

Oh!  I sure am learning a lot today.  Thanks for the description.

Cheers
	Nick

Re: New ABI demangler support in binutils

[Chris Faylor]

On Thu, Jun 15, 2000 at 12:51:30PM -0700, Nick Clifton wrote:
>Hi Chris,
>
>: It probably goes without saying, but I'll ask it anyway -- when you copy
>: libiberty you also will have to move include files from the top level
>: include directory.  Should the include directory be moved over
>: completely as well or should only specific files be copied.
>
>Really ?  Do you mean all of the files in include/ or just
>include/libiberty.h ?

Here's a crudely derived list:

ansidecl.h
demangle.h
dyn-string.h
floatformat.h
getopt.h
hashtab.h
libiberty.h
objalloc.h
obstack.h
partition.h
sort.h
splay-tree.h

These include files are referenced in the gcc 'libiberty' and also
present in 'include'.

cgf

Re: New ABI demangler support in binutils

[Ian Lance Taylor]

   Date: Thu, 15 Jun 2000 12:51:30 -0700
   From: Nick Clifton <nickc@cygnus.com>

   Really ?  Do you mean all of the files in include/ or just
   include/libiberty.h ?

   If it is the latter then I understand, but if it is the former, then I
   am concerned.  surely libiberty only needs its own header file ?

There are a number of files in the include directory which are
associated with libiberty.  libiberty.h is only one of them.

In general, the entire top level of the include directory should be
the same in gcc and the binutils.

Ian

Re: New ABI demangler support in binutils

[Nick Clifton]

Hi Chris,

: It probably goes without saying, but I'll ask it anyway -- when you copy
: libiberty you also will have to move include files from the top level
: include directory.  Should the include directory be moved over
: completely as well or should only specific files be copied.

Really ?  Do you mean all of the files in include/ or just
include/libiberty.h ?

If it is the latter then I understand, but if it is the former, then I
am concerned.  surely libiberty only needs its own header file ?

Cheers
	Nick

Re: New ABI demangler support in binutils

[Nick Clifton]

Hi Mark,

: Nick, these changes should be very safe.  There are almost no
: modifications to the old demangler; there's just a brand-new
: additional source file.

Great!  that makes me feel a lot better.

Cheers
	Nick

Re: New ABI demangler support in binutils

[H . J . Lu]

On Thu, Jun 15, 2000 at 12:19:55PM -0400, Ken Block wrote:
> The demangler support for the new API is available in the libiberty that
> comes with GCC, but does not seem to be available in binutils. How does
> one go about getting demangler support for the new ABI in binutils? How
> is libiberty kept in sync between GCC, gdb, and binutils?
> 

Someone has to sync it. Nick, should I give it a try?


H.J.

New ABI demangler support in binutils

[Ken Block]

The demangler support for the new API is available in the libiberty that
comes with GCC, but does not seem to be available in binutils. How does
one go about getting demangler support for the new ABI in binutils? How
is libiberty kept in sync between GCC, gdb, and binutils?