fuzz_opt fail due to parsing / validating issue in wasm2c

[MaxGraey]

wasm2c b.wasm -o wasm.c
000000d: error: unexpected type form (got 0x4f)
!
-----------------------------------------
Exception:
  File "./scripts/fuzz_opt.py", line 1322, in <module>
  File "./scripts/fuzz_opt.py", line 1118, in test_one
  File "./scripts/fuzz_opt.py", line 775, in handle_pair
  File "./scripts/fuzz_opt.py", line 784, in run_vms
  File "./scripts/fuzz_opt.py", line 705, in run
  File "./scripts/fuzz_opt.py", line 80, in run
  File "/Applications/Xcode.app/Contents/Developer/Library/Frameworks/Python3.framework/Versions/3.8/lib/python3.8/subprocess.py", line 415, in check_output
    return run(*popenargs, stdout=PIPE, timeout=timeout, check=True,
  File "/Applications/Xcode.app/Contents/Developer/Library/Frameworks/Python3.framework/Versions/3.8/lib/python3.8/subprocess.py", line 516, in run
    raise CalledProcessError(retcode, process.args,
-----------------------------------------
!
1
['wasm2c', 'b.wasm', '-o', 'wasm.c']

[MaxGraey]

Example seed: 12980572199263807178

randomized feature opts: 
--all-features 
--nominal 
--disable-threads 
--disable-mutable-globals 
--disable-nontrapping-float-to-int 
--disable-simd 
--disable-bulk-memory 
--disable-sign-ext 
--disable-exception-handling 
--disable-tail-call 
--disable-reference-types 
--disable-multivalue 
--disable-gc 
--disable-memory64 
--disable-typed-function-references 
--disable-relaxed-simd 
--disable-extended-const

wasm-opt run with:

binaryen/bin/wasm-opt b.wasm --emit-wasm2c-wrapper=main.c --nominal --strip-dwarf
warning: no output file specified, not emitting output
wasm2c b.wasm -o wasm.c
000000d: error: unexpected type form (got 0x4f)

[MaxGraey]

@kripken Looking or original.wasm for 12980572199263807178 I see all type declarations uses func_subtype attribute which relate to nominal type system. Is it ok for non-GC feature? Or it's only in text variant and doesn't affect to binary somehow?

(module
 (type $none_=>_none (func_subtype func))
 (type $none_=>_i32 (func_subtype (result i32) func))
 (type $f64_=>_none (func_subtype (param f64) func))
 (type $f64_=>_f64 (func_subtype (param f64) (result f64) func))
 (type $none_=>_i64 (func_subtype (result i64) func))
 (type $i32_=>_none (func_subtype (param i32) func))
 (type $i64_=>_none (func_subtype (param i64) func))
 (type $f32_=>_none (func_subtype (param f32) func))
 (type $none_=>_f32 (func_subtype (result f32) func))
 (type $f64_=>_f32 (func_subtype (param f64) (result f32) func))
 (type $f64_f32_f64_=>_f32 (func_subtype (param f64 f32 f64) (result f32) func))
 (type $i32_f32_i64_=>_i32 (func_subtype (param i32 f32 i64) (result i32) func))
 (type $i64_f32_=>_f64 (func_subtype (param i64 f32) (result f64) func))
 (type $i32_f64_f32_=>_f32 (func_subtype (param i32 f64 f32) (result f32) func))
 (type $none_=>_f64 (func_subtype (result f64) func))
 (type $f32_=>_f64 (func_subtype (param f32) (result f64) func))
 (type $f64_i32_i32_f64_f64_=>_f32 (func_subtype (param f64 i32 i32 f64 f64) (result f32) func))
 (type $f64_f32_i32_f64_=>_i64 (func_subtype (param f64 f32 i32 f64) (result i64) func))
 (type $i64_=>_i32 (func_subtype (param i64) (result i32) func))
 (type $i64_f32_f32_=>_f32 (func_subtype (param i64 f32 f32) (result f32) func))
 (type $f64_f64_f64_=>_i64 (func_subtype (param f64 f64 f64) (result i64) func))
 (type $i64_i32_f32_f64_=>_f32 (func_subtype (param i64 i32 f32 f64) (result f32) func))
 (type $i64_f64_i32_f64_f64_i64_=>_f64 (func_subtype (param i64 f64 i32 f64 f64 i64) (result f64) func))
 (type $f32_f64_=>_i32 (func_subtype (param f32 f64) (result i32) func))
 (type $f32_f32_=>_none (func_subtype (param f32 f32) func))
 (type $f32_=>_f32 (func_subtype (param f32) (result f32) func))

[kripken]

I think func_subtype is fine. But I'll debug this now and see.

[kripken]

@tlively What seems to be the problem here is that the fuzzer emits testcases using --nominal --disable-gc, and those files use a single recursion group for the function types, even though GC is disabled. Trying to load those files in wabt errors on it not recognizing recursion groups.

Should we not emit recursion groups with GC is disabled?

[tlively]

Yes, that sounds like the right fix 👍

[tlively]

I'll have a PR up for this shortly.