Support enabling debug symbol generation for wasm

[tschneidereit]

Viceroy currently doesn't enable, or support enabling, generation of wasm debug symbols, because of an issue with Wasmtime.

It'd be great if we could figure out a way to fix this, so we can support using native debuggers on services' code.

[peterhuene]

Is there more context into the Wasmtime issue already somewhere? A backtrace of the hang would be extremely useful.

[pchickey]

I ran into this issue and left the comment a while ago while on an unrelated adventure and never dug in.

Right now, if you enable it, make test fails with:

     Running tests/body.rs (target/debug/deps/body-11984d2c1fa2bd56)

running 2 tests
test bodies_can_be_written_and_read ... FAILED
test bodies_can_be_written_and_appended ... FAILED

failures:

---- bodies_can_be_written_and_read stdout ----
thread 'bodies_can_be_written_and_read' panicked at 'failed to set up execution context: Other(Debug information error

Caused by:
    The end offset of a location list entry must not be before the beginning.)', cli/tests/common.rs:121:14
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace

---- bodies_can_be_written_and_appended stdout ----
thread 'bodies_can_be_written_and_appended' panicked at 'failed to set up execution context: Other(Debug information error

Caused by:
    The end offset of a location list entry must not be before the beginning.)', cli/tests/common.rs:121:14


failures:
    bodies_can_be_written_and_appended
    bodies_can_be_written_and_read

test result: FAILED. 0 passed; 2 failed; 0 ignored; 0 measured; 0 filtered out; finished in 4.87s

error: test failed, to rerun pass '-p viceroy --test body'
make: *** [Makefile:18: test-crates] Error 101

[peterhuene]

I recall this error coming from gimli with a problem relating to the wasm's debug info section; perhaps the generated section isn't valid for some reason?

[peterhuene]

This is the check that is failing in gimli:

https://github.com/gimli-rs/gimli/blob/master/src/read/loclists.rs#L605-L608

Looks like range.begin is 0xfffffffe and range.end is 204 for me. Trying to determine if gimli bug or toolset bug.

[peterhuene]

So it appears one of the location list entries has a base address of 0xfffffffe, which for this offset pair of begin: 0, end: 206 translates to a begin at 0xfffffffe and an end at 204 as it wraps with a 32-bit address, hence the invalid range error.

[peterhuene]

From what I can tell, the .debug_loc section here has a problematic entry 0xffffffff for begin and 0xfffffffe for end, which is being parsed as a "base address selection entry" with a base address of 0xfffffffe for the subsequent entries.

$ dwarfdump write-and-read-body.wasm --debug-loc gives:

.debug_loc contents:
0x00000000: 
            (0x00000000, 0x00000034): DW_OP_WASM_location 0x0 +0, DW_OP_stack_value

0x00000016: 
            [0xfffffffe, 0x100000035): DW_OP_WASM_location 0x0 +1, DW_OP_stack_value

0x00000034: 
            [0xfffffffe, 0x100000063): DW_OP_WASM_location 0x0 +2, DW_OP_stack_value

...

Someone with more DWARF experience than I probably needs to look at this.

[itsrainy]

I just poked at this a bit and it seems to have been fixed since the last discussion here (possibly by bytecodealliance/wasmtime#2901, bytecodealliance/wasmtime#5358, or some other dependency upgrade). Setting debug_info to true and running make test currently succeeds.