DWARF debugging format implementation

DWARF debugging format implementation

[Devang Patel]

I'd like to know the status of implementation of DWARF debugging
format in FSF tool chain. What is the status of C++ namespace
support and duplicate header removal? I see some code for these
features in GCC, but I do not know whether
1) it is incomplete
2) it is waiting for support from GDB and/or binutil
3) 1) & 2)
4) something else

Thank you,
--
Devang

Re: DWARF debugging format implementation

[Daniel Berlin]

On Sep 29, 2003, at 8:42 PM, Devang Patel wrote:

>
> I'd like to know the status of implementation of DWARF debugging
> format in FSF tool chain. What is the status of C++ namespace
> support and duplicate header removal? I see some code for these
> features in GCC, but I do not know whether
> 1) it is incomplete

I had patches to do namespace support, but it required moving some 
stuff like some c++ specific tree codes to the common codes.
It may be easier to do it cleaner now, it may not be.
David Carlton has probably newer veresions of my patches.

> 2) it is waiting for support from GDB and/or binutil

Yes to this for duplicate header removal.
DW_FORM_ref_addr support in particular.

This will require significant changes to the DWARF2 reader.

It should at least ignore DW_TAG_namespace properly now, so ...

> 3) 1) & 2)
> 4) something else
>
> Thank you,
> --
> Devang
>

Re: DWARF debugging format implementation

[David Carlton]

On Mon, 29 Sep 2003 21:17:47 -0400, Daniel Berlin <dberlin@dberlin.org> said:
> On Sep 29, 2003, at 8:42 PM, Devang Patel wrote:

>> I'd like to know the status of implementation of DWARF debugging
>> format in FSF tool chain. What is the status of C++ namespace
>> support and duplicate header removal? I see some code for these
>> features in GCC, but I do not know whether
>> 1) it is incomplete

> I had patches to do namespace support, but it required moving some
> stuff like some c++ specific tree codes to the common codes.
> It may be easier to do it cleaner now, it may not be.
> David Carlton has probably newer veresions of my patches.

The latest versions that I have are in the log for bug 11114.  They
work with 3.2; they don't work with 3.3, I think (I haven't tried).

>> 2) it is waiting for support from GDB and/or binutil

> Yes to this for duplicate header removal.
> DW_FORM_ref_addr support in particular.

> This will require significant changes to the DWARF2 reader.

> It should at least ignore DW_TAG_namespace properly now, so ...

GDB 5.3 and later are safe to use with compilers generating
DW_TAG_namespace.  GDB 6.1 will probably take advantage of that
information (and, in general, will handle nested types better); with a
bit of luck, the relevant patches will hit mainline GDB in the next
month or two.

(I have no idea who, if anybody, is working on duplicate header
removal.)

David Carlton
carlton@kealia.com

Re: DWARF debugging format implementation

[Devang Patel]

David,

It is good to know about C++ namespace support progress. Thanks
for update.

On Tuesday, September 30, 2003, at 8:33 AM, David Carlton wrote:

> (I have no idea who, if anybody, is working on duplicate header
> removal.)

Does this mean, GCC work is complete, but GDB does not know how
to handle it? What about linker ? Does it know how to remove duplicate
header info?

This leads me to another question - is there any document describing
what GCC is generating for duplicate header debug info removal ?

I'll try namespace patch (11114) and see what happens.

Thanks,
--
Devang

Re: DWARF debugging format implementation

[Daniel Berlin]

On Sep 30, 2003, at 2:44 PM, Devang Patel wrote:

> David,
>
> It is good to know about C++ namespace support progress. Thanks
> for update.
>
> On Tuesday, September 30, 2003, at 8:33 AM, David Carlton wrote:
>
>> (I have no idea who, if anybody, is working on duplicate header
>> removal.)
>
> Does this mean, GCC work is complete
yes

>  but GDB does not know how
> to handle it?
yes

> What about linker ? Does it know how to remove duplicate
> header info?
>
It's done through linkonce sections, so it doesn't do anything special.
This, of course, adds some constant overhead that could be removed by 
ld knowing how to remove duplicate info on it's own.
> This leads me to another question - is there any document describing
> what GCC is generating for duplicate header debug info removal ?
>
> I'll try namespace patch (11114) and see what happens.
>

I haven't looked at exactly which version was the last I had sent 
david, but i eventually extended it to support using statements, 
anonymous namespaces, etc.
>

Re: DWARF debugging format implementation

[David Carlton]

On Tue, 30 Sep 2003 14:49:43 -0400, Daniel Berlin <dberlin@dberlin.org> said:

> I haven't looked at exactly which version was the last I had sent
> david, but i eventually extended it to support using statements,
> anonymous namespaces, etc.

I don't believe I ever got a version of your GCC patch which correctly
supported using statements.  I know I have a version which generates
DW_TAG_namespace, which handles anonymous namespaces correctly, and
does one other thing which I haven't yet implemented support for in
GDB.  (Maybe namespace aliases?)

It would be nice if the support for DW_TAG_namespace went into GCC
3.4.

David Carlton
carlton@kealia.com

Re: DWARF debugging format implementation

[Daniel Berlin]

On Sep 30, 2003, at 3:37 PM, David Carlton wrote:

> On Tue, 30 Sep 2003 14:49:43 -0400, Daniel Berlin 
> <dberlin@dberlin.org> said:
>
>> I haven't looked at exactly which version was the last I had sent
>> david, but i eventually extended it to support using statements,
>> anonymous namespaces, etc.
>
> I don't believe I ever got a version of your GCC patch which correctly
> supported using statements.
I don't remember if i ever sent that version to you.
USING_STMT can contain anything, and frequently did (it had OVERLOAD 
trees in them, etc that we shouldn't be seeing in the backend).

>   I know I have a version which generates
> DW_TAG_namespace, which handles anonymous namespaces correctly, and
> does one other thing which I haven't yet implemented support for in
> GDB.  (Maybe namespace aliases?)

That was it, I think.

>
> It would be nice if the support for DW_TAG_namespace went into GCC
> 3.4.
>
It's unlikely that I'll have the time or energy to do this, but i can 
provide help to anyone who wants to try.

Re: DWARF debugging format implementation

[Daniel Jacobowitz]

On Tue, Sep 30, 2003 at 03:41:48PM -0400, Daniel Berlin wrote:
> 
> On Sep 30, 2003, at 3:37 PM, David Carlton wrote:
> 
> >On Tue, 30 Sep 2003 14:49:43 -0400, Daniel Berlin 
> ><dberlin@dberlin.org> said:
> >
> >>I haven't looked at exactly which version was the last I had sent
> >>david, but i eventually extended it to support using statements,
> >>anonymous namespaces, etc.
> >
> >I don't believe I ever got a version of your GCC patch which correctly
> >supported using statements.
> I don't remember if i ever sent that version to you.
> USING_STMT can contain anything, and frequently did (it had OVERLOAD 
> trees in them, etc that we shouldn't be seeing in the backend).
> 
> >  I know I have a version which generates
> >DW_TAG_namespace, which handles anonymous namespaces correctly, and
> >does one other thing which I haven't yet implemented support for in
> >GDB.  (Maybe namespace aliases?)
> 
> That was it, I think.
> 
> >
> >It would be nice if the support for DW_TAG_namespace went into GCC
> >3.4.
> >
> It's unlikely that I'll have the time or energy to do this, but i can 
> provide help to anyone who wants to try.

If one of you will figure out who has the newest version and send it to
me, I'll give it a shot.

-- 
Daniel Jacobowitz
MontaVista Software                         Debian GNU/Linux Developer

Re: DWARF debugging format implementation

[David Carlton]

On Tue, 30 Sep 2003 17:37:53 -0400, Daniel Jacobowitz <drow@mvista.com> said:
> On Tue, Sep 30, 2003 at 03:41:48PM -0400, Daniel Berlin wrote:
>> On Sep 30, 2003, at 3:37 PM, David Carlton wrote:

>>> It would be nice if the support for DW_TAG_namespace went into GCC
>>> 3.4.

>> It's unlikely that I'll have the time or energy to do this, but i
>> can provide help to anyone who wants to try.

> If one of you will figure out who has the newest version and send it
> to me, I'll give it a shot.

I've just attached what I've been using to bug 11114.  It's not the
best patch in the world - its way of detecting anonymous namespaces is
gross - but it seems to work.

I don't know the status of Daniel's patches: the patch I attached is
an early patch of his + anonymous namespace fix + context die fix, but
I know he's updated his patch since then.

David Carlton
carlton@kealia.com

Re: DWARF debugging format implementation

[Daniel Berlin]

On Sep 30, 2003, at 5:55 PM, David Carlton wrote:

> On Tue, 30 Sep 2003 17:37:53 -0400, Daniel Jacobowitz 
> <drow@mvista.com> said:
>> On Tue, Sep 30, 2003 at 03:41:48PM -0400, Daniel Berlin wrote:
>>> On Sep 30, 2003, at 3:37 PM, David Carlton wrote:
>
>>>> It would be nice if the support for DW_TAG_namespace went into GCC
>>>> 3.4.
>
>>> It's unlikely that I'll have the time or energy to do this, but i
>>> can provide help to anyone who wants to try.
>
>> If one of you will figure out who has the newest version and send it
>> to me, I'll give it a shot.
>
> I've just attached what I've been using to bug 11114.  It's not the
> best patch in the world - its way of detecting anonymous namespaces is
> gross - but it seems to work.
>
Yeah, i did it differently later on.
> I don't know the status of Daniel's patches: the patch I attached is
> an early patch of his + anonymous namespace fix + context die fix, but
> I know he's updated his patch since then.

I have, but i doubt i have it still. Time for a huge grep session :P

I haven't looked at this stuff in quite a while.

>
> David Carlton
> carlton@kealia.com