On 2011-11-22 21:04, bioinfornatics wrote:
> Le mardi 22 novembre 2011 à 16:27 +0100, Jacob Carlborg a écrit :
>> On 2011-11-22 16:19, Martin Nowak wrote:
>>> - Are there other DWARF debuggers that we should take into account?
>>
>> The LLBD debugger: http://lldb.llvm.org/
>>
>
> LLDB seem to be great but is OSX only ;'(

I'm certain it will be available on other platforms as well. This page mentions something about Linux: http://lldb.llvm.org/build.html

-- 
/Jacob Carlborg

On Tue, 22 Nov 2011 12:19:00 -0300, Martin Nowak <dawg@dawgfoto.de> wrote:

>   - Are there other DWARF debuggers that we should take into account?

Depends on what "taking into account" means in this context.

The OpenWatcom debugger (wd/wdw) supports DWARF-2, and is released for
WIN32, OS/2 and Linux (wd only). It (wd) probably compiles on BSD and
OS/X as well.
wd is the full screen character based version.
wdw is the gui version.

http://www.openwatcom.org

Roald

On 2011-11-23 13:24, Roald Ribe wrote:
> On Tue, 22 Nov 2011 12:19:00 -0300, Martin Nowak <dawg@dawgfoto.de> wrote:
>
>> - Are there other DWARF debuggers that we should take into account?
>
> Depends on what "taking into account" means in this context.
>
> The OpenWatcom debugger (wd/wdw) supports DWARF-2, and is released for
> WIN32, OS/2 and Linux (wd only). It (wd) probably compiles on BSD and
> OS/X as well.
> wd is the full screen character based version.
> wdw is the gui version.
>
> http://www.openwatcom.org
>
> Roald

For Mac OS X it would probably need support for the Mach-O file format, if it already isn't supported.

-- 
/Jacob Carlborg

On 11/22/2011 7:19 AM, Martin Nowak wrote:
> - Can we slowly start to acquire DWARF-3/4?

Dwarf 2 is good enough, so why do 3/4?

> - Why were extensions chosen over say representing an array as two field struct?

Because it is presumably harder to get a debugger to recognize a magic struct than explicitly give it a type.

> - There is this request by Robert Clipsham
> http://dwarfstd.org/ShowIssue.php?issue=100504.1.
> What's the state of this, it wouldn't be needed any longer.
>
> - We should contact http://www.zero-bugs.com, it is said to support the
> dwarf extensions and would need to be changed.
>
> - Find out who else is affected by changing the current extensions.
>
> - Fixing the D-ABI specification.

Are the issues in bugzilla?

> - Are there other DWARF debuggers that we should take into account?
>
> - Adding a page to GNU wiki describing the D extensions.
>
> - We could probably use DW_TAG_GNU_template_parameter_pack for variadic template
> arguments.
> http://wiki.dwarfstd.org/index.php?title=C%2B%2B0x:_Variadic_templates
>
> martin

Martin Nowak Wrote:

> I think we should follow Tom Tromney's proposal to add the extensions under the GNU vendor space, i.e. prefix them with 0x41xx.

Yeah, why use extensible debug info format and not use the extensibility feature?

"Martin Nowak" <dawg@dawgfoto.de> writes:

> On Tue, 22 Nov 2011 04:18:47 +0100, Andrei Alexandrescu <SeeWebsiteForEmail@erdani.org> wrote:
>
>> Maybe someone knowledgeable could chime in:
>>
>> http://sourceware.org/ml/gdb/2011-11/msg00066.html
>>
>>
>>
>> Andrei
>
> I think we should follow Tom Tromney's proposal to add the extensions
> under the GNU vendor space, i.e. prefix them with 0x41xx.
> Numbers hopefully don't fall under licensing issues.
>
> It should be fairly straightforward to improve gdb debugging support
> given that
> changes are accepted on either side. The GDB codebase is a slightly
> unfriendly
> environment but getting rid of 'p ((char*[2])str)[1]' would be great.

I've done a bit of reading on this issue, at least with regards to D arrays.  I'm not a DWARF expert, but it appears that D dynamic arrays could be represented using the existing DWARF support, without an extension.

Associative arrays are another issue.  The current ABI describes them an opaque type.  However, for the debugger to work with them, the structure must be understood somehow by the debugger.  If treating them as a void* is the right approach (it still will result in compiler incompatibility), the ABI must define functions that access AA's.  And debuggers will have to call those functions to read the AA.  Until then, the debugger would have to cheat to access AA's.

I don't know enough about delegates to comment on the support issues there.

> I would volunteer to do the implementation but it would be great if
> someone helped with
> clarifying the following issues.
>
>  - Can we slowly start to acquire DWARF-3/4?
>
>  - Why were extensions chosen over say representing an array as two
> field struct?

There was a comment on the ABI page that a struct is passed to functions differently from an array, and therefore debuggers wouldn't handle D arrays correctly if described that way.

>  - We could probably use DW_TAG_GNU_template_parameter_pack for
> variadic template arguments.
>    http://wiki.dwarfstd.org/index.php?title=C%2B%2B0x:_Variadic_templates

Note that the C++-11 standard also provides variadic templates. Therefore DWARF probably needs to address this in some standard way if it doesn't already.  I'd think that D and C++ could share the same solution.

Walter Bright <newshound2@digitalmars.com> writes:

> On 11/22/2011 7:19 AM, Martin Nowak wrote:
>> - Can we slowly start to acquire DWARF-3/4?
>
> Dwarf 2 is good enough, so why do 3/4?
>
>> - Why were extensions chosen over say representing an array as two field struct?
>
> Because it is presumably harder to get a debugger to recognize a magic struct than explicitly give it a type.
>
>> - There is this request by Robert Clipsham
>> http://dwarfstd.org/ShowIssue.php?issue=100504.1.
>> What's the state of this, it wouldn't be needed any longer.
>>
>> - We should contact http://www.zero-bugs.com, it is said to support the dwarf extensions and would need to be changed.
>>
>> - Find out who else is affected by changing the current extensions.
>>
>> - Fixing the D-ABI specification.
>
> Are the issues in bugzilla?

Huh, I swore I had already filed a bug regarding the AA ABI.

http://d.puremagic.com/issues/show_bug.cgi?id=7008

Jerry

On Fri, 25 Nov 2011 04:13:18 +0100, Walter Bright <newshound2@digitalmars.com> wrote:

> On 11/22/2011 7:19 AM, Martin Nowak wrote:
>> - Can we slowly start to acquire DWARF-3/4?
>
> Dwarf 2 is good enough, so why do 3/4?
>
>> - Why were extensions chosen over say representing an array as two field struct?
>
> Because it is presumably harder to get a debugger to recognize a magic struct than explicitly give it a type.
>
Dwarf-4 supports arbitrary expressions for array base addresses and the upper bound,
one could directly translate D arrays. GDB doesn't support it at the moment,
but it basically rules out requesting an extension to the DWARF standard.

I think currently the best decision is to use the existing D support in GDB.

C++0x will likely lead to extended support in GDB, while llvm-db should
improve the OSX debugger support.
With better support for the new formats we should be able to express specific D
constructs in DWARF.

>> - There is this request by Robert Clipsham
>> http://dwarfstd.org/ShowIssue.php?issue=100504.1.
>> What's the state of this, it wouldn't be needed any longer.
>>
>> - We should contact http://www.zero-bugs.com, it is said to support the
>> dwarf extensions and would need to be changed.
>>
>> - Find out who else is affected by changing the current extensions.
>>
>> - Fixing the D-ABI specification.
>
> Are the issues in bugzilla?
>
>> - Are there other DWARF debuggers that we should take into account?
>>
>> - Adding a page to GNU wiki describing the D extensions.
>>
>> - We could probably use DW_TAG_GNU_template_parameter_pack for variadic template
>> arguments.
>> http://wiki.dwarfstd.org/index.php?title=C%2B%2B0x:_Variadic_templates
>>
>> martin