Query regarding new dwarf type support in gdb

Query regarding new dwarf type support in gdb

[Saraswati, Sujoy (JCTL-MCBS)]

Hi all,
 I have a question on new dwarf type supports in gdb. For example, http://dwarfstd.org/ShowIssue.php?issue=090106.1&type=closed3 talks about a new DWARF tag DW_TAG_rvalue_reference_type to be added for debugging C++0x R-Value references. My question is, when the compilers start emitting this new tag in the debug sections, how does the older gdb handle it ? Does it gracefully skip this unknown type, or does it error out ? The dwarf2read.c file in gdb has many instances when error() is called. My understanding is that if gdb calls error() while processing the dwarf information for a module, the dwarf processing would be incomplete and debugging might get affected.
Regards,
Sujoy

Re: Query regarding new dwarf type support in gdb

[Tom Tromey]

>>>>> "Sujoy" == Saraswati, Sujoy (JCTL-MCBS) <sujoy.saraswati@hp.com> writes:

Sujoy>  I have a question on new dwarf type supports in gdb. For example,
Sujoy> http://dwarfstd.org/ShowIssue.php?issue=090106.1&type=closed3 talks
Sujoy> about a new DWARF tag DW_TAG_rvalue_reference_type to be added for
Sujoy> debugging C++0x R-Value references. My question is, when the compilers
Sujoy> start emitting this new tag in the debug sections, how does the older
Sujoy> gdb handle it ? Does it gracefully skip this unknown type, or does it
Sujoy> error out ? The dwarf2read.c file in gdb has many instances when
Sujoy> error() is called. My understanding is that if gdb calls error() while
Sujoy> processing the dwarf information for a module, the dwarf processing
Sujoy> would be incomplete and debugging might get affected.

I think gdb should issue a complaint (try "set complaints 1000") and
then make a new TYPE_CODE_ERROR type.

IIRC, g++ already emis rvalue references, but gdb doesn't handle them yet.
You can see this in action by digging around in libstdc++, or maybe
using gdb with complaints enabled and "file -readnow ...".

We try to avoid calling 'error' during DWARF reading (not with complete
success) since it tends to discard the whole file's debuginfo -- much
too big a hammer.  This area could use some fixes.

Tom

RE: Query regarding new dwarf type support in gdb

[Saraswati, Sujoy (JCTL-MCBS)]

Hi Tom,
  Thanks for your reply. I have some further comment below -

> I think gdb should issue a complaint (try "set complaints 1000") and
> then make a new TYPE_CODE_ERROR type.
> 
> IIRC, g++ already emis rvalue references, but gdb doesn't handle them
> yet.
> You can see this in action by digging around in libstdc++, or maybe
> using gdb with complaints enabled and "file -readnow ...".

Ok, I could check the behavior. In general, is there any place where I can check for current debug support of FSF gdb for C++11 features ?

> We try to avoid calling 'error' during DWARF reading (not with complete
> success) since it tends to discard the whole file's debuginfo -- much
> too big a hammer.  This area could use some fixes.

Yes, I agree. It would be good if we could continue reading the rest of the debug information from the file even if we face an unknown type.

Regards,
Sujoy

> Tom

Re: Query regarding new dwarf type support in gdb

[Tom Tromey]

>>>>> "Sujoy" == Saraswati, Sujoy (JCTL-MCBS) <sujoy.saraswati@hp.com> writes:

Sujoy> Ok, I could check the behavior. In general, is there any place where I
Sujoy> can check for current debug support of FSF gdb for C++11 features ?

Just searching around in gdb/dwarf2read.c; and somtimes other dwarf*.c
files.

Sujoy> Yes, I agree. It would be good if we could continue reading the rest
Sujoy> of the debug information from the file even if we face an unknown
Sujoy> type.

I think this specific case should work ok.

Tom