elfutils use patterns in frysk

elfutils use patterns in frysk

[Sami Wagiaalla]

Hi Frank,

A while back you asked about elfutils use patterns in Frysk.
I only know abut libdw* so everybody feel free to add to this:

Okay so there are two types of items here:

Things we do routinely in frysk that require more than one libdw* call:

- getting the function die corresponding an address:
   - dwfl_addrdie (get compilation unit)
   - dwarf_get_scopes (first scope is the narrowest die containing the 
give address)
   - dwarf_get_scopes_die (get physical scopes)
   - iterate until the first DW_TAG_subprogram
- Getting functions have have been inlined into a function currently in 
frame
  simmilar to above.


Things we do to get C++ information out of the C'ish dwarf information 
(also involves more than one libdw* function call, hurestics, etc.):

- Getting a class to which a function belongs
   similar to getting functions above, with an extra step to follow 
DW_TAG_specification
- Deciding when a member function/variable is/isnt static
  - checking Tags and artificial parameter.

I cant think of more atm... I will update you as I do.
If there isnt enough detail in the above please let me know.

Cheers,
  Sami

Re: elfutils use patterns in frysk

[Phil Muldoon]

Sami Wagiaalla wrote:
> Hi Frank,
>
> A while back you asked about elfutils use patterns in Frysk.
> I only know abut libdw* so everybody feel free to add to this:

FWIW I believe the current code that is in the frysk-sys/lib/dwfl is 
based on a simple facade OOP pattern (as far as I can tell, I was not 
the original author of this code). So imo it should be noted that what 
we have are not, and should never be considered as a binding.  For 
example there is no memory management, and there is a simple 
transactional contract that does not exist outside of each call.

>
> Okay so there are two types of items here:

A third usage pattern is elf bindings for building corefiles, which are 
not dwfl based. These are fundamental bindings (ElfSection, ElfEHeader, 
ElfSectionHeader, ElfProgramHeader, ElfData) and so on.These are 
basically 1:1 mappings on the Gelf api. There is also some meta-code 
that deals with corefile note based bindings, but these are likely to be 
very specialized. I do not consider these as a first priority (or a 
second, or third) for Frank's work. I can't think who would use them 
other than Frysk and we already have custom code that deals with that.

I know Petr uses these in some way as well. Maybe he can add more detail 
here.

Regards

Phil

Re: elfutils use patterns in frysk

[Petr Machata]

[-- Attachment #1: Type: text/plain, Size: 610 bytes --]

Phil Muldoon wrote:
> A third usage pattern is elf bindings for building corefiles, which are
> not dwfl based. These are fundamental bindings (ElfSection, ElfEHeader,
> ElfSectionHeader, ElfProgramHeader, ElfData) and so on.These are
> basically 1:1 mappings on the Gelf api.

[...]

> I know Petr uses these in some way as well. Maybe he can add more detail
> here.

I don't really deal with corefiles.  I've written what I needed for
fl?trace, which was handling of symbols and their versions, and few
related things.  Those are usually very straightforward libelf wrappers.

> Phil

PM


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 189 bytes --]

Re: elfutils use patterns in frysk

[Phil Muldoon]

Petr Machata wrote:
> I don't really deal with corefiles.  I've written what I needed for
> fl?trace, which was handling of symbols and their versions, and few
> related things.  Those are usually very straightforward libelf wrappers.
>
>   
Sorry, not in respect to corefiles, but what you have written for symbol 
usage. I think general solicitation as to how Frysk uses elfutils would 
build a bigger picture so that the C++ bindings Frank is working on 
could be properly targeted?

Regards

Phil
>> Phil
>>     
>
> PM
>
>