How to get a call stack when wasm-ctor-eval stops?

[rdb]

Hello,

Apologies if this is already documented somewhere, but I was unable to find it. I'm trying to find out how I can easily find where a particular call to an import is being made when wasm-ctor-eval hits something it can't eval. I was hoping to find some debug option so that I could see something like a traceback.

At the moment I am manually searching through the wasm disassembly, which is quite a painstaking process given that this is a huge binary. Any suggestions for making this process easier would be very helpful.

[kripken]

We don't have an option for this atm. You can approximate it by running in a debugger and stopping on any FailToEvalException, then inspecting the stack. We eval in an interpreter that will have a stack frame for each function called, and includes the function name in the arguments, so the information is there (at least in a debug build).

Supporting this "properly" would require a shadow stack of some sort. Adding a basic version, that notes functions as it enters and leaves them, would at least give a basic stack trace, and shouldn't be too hard to add. Adding precise binary offsets would be possible but harder.

[rdb]

Even just the function names would be very helpful already.

Would it be easier to ask for a spammy debug log feature that just prints an indented name of the function as it enters it? Then I can just work upward from the last bit before it stopped.

[kripken]

Oh, we do have a spammy option already I believe. Add

#define WASM_INTERPRETER_DEBUG 1

at the top of src/wasm-interpreter.h.

[tlively]

The new interpreter will make it much easier to support this use case.

[rdb]

Thanks! I'm looking forward to the new interpreter. I've managed to achieve what I need using wizer in the meantime, which gives a stack trace when it encounters an external call. On top of that, wizer lets me pass in external data (via a wasi call), which I need for my use case while there does not seem to be a way to do this with wasm-ctor-eval.

Are the plans for the new interpreter documented somewhere?

[tlively]

The only documentation so far is this PR/commit description: #7227