Author here, a bit cringe to see your WIP project posted here.
My motivation for writing this from scratch is to simultaneously understand DWARF 5 in all its gory details, develop tooling to support my work on the OCaml compiler, and to build a source debugger in OCaml. The performance or lack of isn't a focus right now, I'm sure it doesn't compare to gimli-rs for performance but it does fully support DWARF 5 which gimli-rs doesn't (missing debug_info section support https://github.com/gimli-rs/gimli/pull/807).
Currently I'm working on the read support, and various bits of tooling around understanding Call Frame Information and simulating the DWARF expression evaluation. I'm unhappy with the tooling GDB/LLDB provide for displaying and debugging this information, I've personally spent far too much time staring at CFI expressions and walking memory looking for where it goes wrong. In the OCaml compiler we have few people that understand or work on this area.
DWARF is such a neat little format. I have a soft spot for the "line program" table, which is a little state machine that allows DWARF to efficiently encode a line/column correspondence for virtual addresses.
One interesting thing about the dwarf format is that it is executable and Turing complete, running small programs on a virtual machine to figure out file names, line numbers and exception handling.
It's really fascinating the different little evaluators that exist in DWARF. I started writing one for the DWARF expression format (https://github.com/tmcgilchrist/durin/pull/7) that I'm integrating with a debugger and free-monad/effect handler style code to lookup the missing data to provide for the Requires* callbacks.
Then there is the line number evaluation, which I haven't started on.
Building a Debugger by Sy Brand (https://nostarch.com/building-a-debugger) is an awesome book if you want to learn more about this. It covers DWARF 4 and you build a full debugger in C++17 for Linux/x86_64. Can't recommend that book highly enough!
It is heavily WIP at the moment so I haven't published to opam just yet. You'll need to compile from source if you want to try it out. I'm filling in examples as I need them and publishing code when it isn't too terrible to look at.
How does this compare with gimli-rs? Last I checked it was the absolute leader in performance.
https://github.com/gimli-rs/gimli
Author here, a bit cringe to see your WIP project posted here.
My motivation for writing this from scratch is to simultaneously understand DWARF 5 in all its gory details, develop tooling to support my work on the OCaml compiler, and to build a source debugger in OCaml. The performance or lack of isn't a focus right now, I'm sure it doesn't compare to gimli-rs for performance but it does fully support DWARF 5 which gimli-rs doesn't (missing debug_info section support https://github.com/gimli-rs/gimli/pull/807).
Currently I'm working on the read support, and various bits of tooling around understanding Call Frame Information and simulating the DWARF expression evaluation. I'm unhappy with the tooling GDB/LLDB provide for displaying and debugging this information, I've personally spent far too much time staring at CFI expressions and walking memory looking for where it goes wrong. In the OCaml compiler we have few people that understand or work on this area.
> Author here, a bit cringe to see your WIP project posted here.
I'm confused; are you referring to GP's link to `gimli`, or your own repo in the OP?
My own repo in the OP :-)
I sure hope there’s an anti-debug framework called Durin’s Bane?
Bad idea to delve too deep into the binary.
"The dwarves debugged too greedily and too deep. You know what they awoke in the darkness of GDB-dum... stack corruption and flame.“
"The dark fire will not avail you, flame of Unix. Go back to the Segfault! You cannot crash."
I hope if you use it in late Ocotober when the moon is out, it opens the CD tray... for those who still have one.
DWARF is such a neat little format. I have a soft spot for the "line program" table, which is a little state machine that allows DWARF to efficiently encode a line/column correspondence for virtual addresses.
Surely, you are joking about the "little" part.
The Dwarf 5 document is over 400 pages long (c.f. https://dwarfstd.org/doc/DWARF5.pdf).
Compared to the languages that it models, that's very small!
One interesting thing about the dwarf format is that it is executable and Turing complete, running small programs on a virtual machine to figure out file names, line numbers and exception handling.
It's really fascinating the different little evaluators that exist in DWARF. I started writing one for the DWARF expression format (https://github.com/tmcgilchrist/durin/pull/7) that I'm integrating with a debugger and free-monad/effect handler style code to lookup the missing data to provide for the Requires* callbacks.
Then there is the line number evaluation, which I haven't started on.
Building a Debugger by Sy Brand (https://nostarch.com/building-a-debugger) is an awesome book if you want to learn more about this. It covers DWARF 4 and you build a full debugger in C++17 for Linux/x86_64. Can't recommend that book highly enough!
And for the misuse angle there is this talk from Shmoocon 2011: https://www.youtube.com/watch?v=nLH7ytOTYto
Several of the examples seem to be empty (e.g. https://github.com/tmcgilchrist/durin/blob/master/example/dw...)
Also, the link to OPAM documentation leads to a 404
It is heavily WIP at the moment so I haven't published to opam just yet. You'll need to compile from source if you want to try it out. I'm filling in examples as I need them and publishing code when it isn't too terrible to look at.
Ah, no worries. I'm just now realizing that you weren't even the one who submitted this to HN.
Best of luck to you! I spent many hours trying to understand and generate proper Dwarf tables.
Are ELF and DWARF names chosen deliberately or was it a coincidence? Lol