Interop broken on Linux?

[guregu]

$ WASMTIME_BACKTRACE_DETAILS=1 go test -v ./trealla   
=== RUN   TestInterop/custom_function
2023/07/19 18:29:51 query: interop_test(X).
2023/07/19 18:29:51 query: X is 1 + 1.
=== NAME  TestInterop
    interop_test.go:34: trealla: query error: error while executing at wasm backtrace:
            0: <unknown>!pl_query
            1: 0x1fe1db - fn_sys_host_call_2
                            at /Users/guregu/code/trealla/fork/src/predicates.c:7692:12
            2: 0x234bc6 - start
                            at /Users/guregu/code/trealla/fork/src/query.c:1611:14
            3: 0x23b115 - execute
                            at /Users/guregu/code/trealla/fork/src/query.c:1823:9
            4: 0xf8327 - run
                            at /Users/guregu/code/trealla/fork/src/parser.c:3630:3
            5: 0x223e56 - pl_query
                            at /Users/guregu/code/trealla/fork/src/prolog.c:162:12
        
        Caused by:
            wasm trap: call stack exhausted

Works ok on macOS+arm64 so probably something weird going on. This is running in a VM though.

[guregu]

Hmm, looks like there's a 1MB (maybe 2MB?) stack size limit by default: bytecodealliance/wasmtime#900
This would explain a lot... the Trealla wasm build uses a 8MB stack.
There's no way to set this size yet for wasmtime-go so I will add it then fix this.
Pretty surprising it doesn't read the stack size from the binary.

[guregu]

This should take care of it: bytecodealliance/wasmtime-go#180

[guregu]

Hmm... it looks like it's more complex than I thought.
wasmtime divides the stack space every time a guest -> host -> guest (reentrant?) call happens
Even with the larger stack, if there's more than 2 or so reentrant queries happening the stack just becomes too small.
This can be worked around by calling Clone() on the instance passed to the native predicate, but incurs an overhead.
It might be possible to avoid this with a more complex locking system but probably not worth it.