DWARF types information

[radare]

The DWARF type information must be loaded in the same sdb database used by cparse. both source formats (dwarf and .c/.h files) must provide the same (or almost the same) information so it can be merged in a single place to be used by structs parsing or variable types.

[montekki]

ok, time to have a discussion about this. how is info stored now and how is it used?
types in dwarf are tree-like structures, how do we store them in sdb?

[radare]

We should reuse and extend what's already done with cparse sdb storage. The idea is that cparse and dwarf (and future pdb) use the same database structure so you can use .h, or other kind of human readable forms to read or serialize type information.

I guess dwarf have more type info than plain C, so we should check that

On 05 May 2014, at 12:20, Fedor Sakharov [email protected] wrote:

ok, time to have a discussion about this. how is info stored now and how is it used?
types in dwarf are tree-like structures, how do we store them in sdb?

—
Reply to this email directly or view it on GitHub.

[XVilka]

Related to your task, @oddcoder

[radare]

@oddcoder pls

[radare]

OMG THIS IS NOT GONNA HAPPEN EVER. UNASSIGNING ODDCODER AND MOVING FORWARD

[radare]

@jpenalbae you are the one intested on this here. maybe you can find some time to do it or push someone to do it

[MaskRay]

Oh DWARF is underappreciated... Any idea to make the current type system more DWARF-friendly and incorporate more information from DWARF?

DWARF 5 introduces a bunch of tags and attributes for type support DW_TAG_formal_parameter DW_TAG_call_site ...

[XVilka]

I think we need to better define our current format of the types storage.
Currently it's stored in types SDB namespace, but I think the exact format of SDB is not efficient for more metainformation. So parsing DWARF just should insert proper SDB records, thats it.

[Maijin]

Have you tried the dwarf r2-extras plugin?

[XVilka]

@MaskRay my current plan (after r2con) is to port some useful changes from the mainstream tcc to our code, I already know what and how to port, to improve parsing of the C structs/types, then improve the SDB storage layout. We can discuss the exact layout in just couple days during the hackaton.

[radare]

I think we should not import some information, just provide some getters. Dwarf format is complex as hell but provides a compressed and structured format to save this info. And importing results in lot of cpu usage for importing into sdb and then more memory used because of this

…

On 4 Sep 2017, at 10:27, Anton Kochkov ***@***.***> wrote: @MaskRay my current plan (after r2con) is to port some useful changes from the mainstream tcc to our code, I already know what and how to port, to improve parsing of the C structs/types, then improve the SDB storage layout. We can discuss the exact layout in just couple days during the hackaton. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or mute the thread.

[Maijin]

dwarf-parser [syslib] dwarf parser
libdwarf [syspkg] libdwarf - git

See https://github.com/radare/radare2-regressions/blob/master/t.extras/dwarf/dwarf please before wanting reiventing what's already been done by @p4n74

[XVilka]

Import should be done, otherwise the whole point of loading DWARF is worth nothing.