DWARF-5 support

[ggreif]

Current Goal

The upcoming milestone is Exprloc support for the Wasm-style locations. I'll try to put together a PR next.

So far I see following problems (possibly related to #1613):

• no operand stack locations yet (middle-term)
• index needs (seems) to begin at 1 (I'd expect 0) to obtain valid ranges
• (potentially resolved by Fix value label ranges resolution #1572) ranges get discarded because not being sufficiently populated with labels.

Next Goal

• accept .debug_types (middle-term)

OUTDATED info below

I gather that the Wasm DWARF spec defines DWARF-5 as the baseline debug format.

However several sections from Version 5 are rejected by wasmtime with panics. E.g.
https://github.com/bytecodealliance/wasmtime/blob/master/crates/debug/src/read_debuginfo.rs#L63

For my purposes ".debug_rnglists" is the most important.

Is there any ETA for supporting these?

Note: The clang-10 compiler (currently in rc1) doesn't reliably emit DWARF-5 yet.

clang-11 (pre-release) seems to generate DWARF-5 without crashing. As per #1410 the DWARF example Wasm files are compiled with this clang.

[yurydelendik]

I gather that the Wasm DWARF spec defines DWARF-5 as the baseline debug format.
For my purposes ".debug_rnglists" is the most important.
Note: The clang-10 compiler (currently in rc1) doesn't reliably emit DWARF-5 yet.

The wasmtime currently consumes LLVM output. As you noticed, that older clang supports DWARF-4, and we are using it as input at the moment.

@ggreif which Wasm code generators you are using that output DWARF 5 sections?

Is there any ETA for supporting these?

I'll keep this issue opened to start tracking missing DWARF 5 features.

[ggreif]

@ggreif which Wasm code generators you are using that output DWARF 5 sections?

It is an in-house compiler. Please p/m me if you need details. I am happy to send generated code to check your ingesters (when you're ready).

[yurydelendik]

@ggreif is anything else left to close this particular issue? I see .debug_types and "operand stack locations". I would suggest to open a separate issue for the latter to track it with https://reviews.llvm.org/D78818.

[ggreif]

I have my own heap of problems, with my compiler. Happy to close this one.