Fast interpreter: issue with decoding block types

[vouillon]

It seems that the fast interpreter can miscompute the length of a block type. I think this happens for multi-byte type indexes. I get this trace for the piece of code below:

jerome@catalpa:~/tmp/bugs$ ~/sources/wizard-engine/bin/spectest.x86-64-linux -ti bug.bin.wast 
##+bug.bin.wast
+1:   block[[] -> [i32 i32]]
+3:   unreachable
##-fail: bug.bin.wast @ 21:2 assert_return expected [], got trap[UNREACHABLE]

(module
  (type (func)) (type (func)) (type (func)) (type (func))
  (type (func)) (type (func)) (type (func)) (type (func))
  (type (func)) (type (func)) (type (func)) (type (func))
  (type (func)) (type (func)) (type (func)) (type (func))
  (type (func)) (type (func)) (type (func)) (type (func))
  (type (func)) (type (func)) (type (func)) (type (func))
  (type (func)) (type (func)) (type (func)) (type (func))
  (type (func)) (type (func)) (type (func)) (type (func))
  (type (func)) (type (func)) (type (func)) (type (func))
  (type (func)) (type (func)) (type (func)) (type (func))
  (type (func)) (type (func)) (type (func)) (type (func))
  (type (func)) (type (func)) (type (func)) (type (func))
  (type (func)) (type (func)) (type (func)) (type (func))
  (type (func)) (type (func)) (type (func)) (type (func))
  (type (func)) (type (func)) (type (func)) (type (func))
  (type (func)) (type (func)) (type (func)) (type (func))
  (type (func))
  (type $res (func (result i32 i32)))
  (func (export "main")
     (block (type $res)
        (return))
     (drop)
     (drop)
  )
)
(assert_return (invoke "main"))

(module definition binary
  "\00\61\73\6d\01\00\00\00\01\c9\81\80\80\00\42\60"
  "\00\00\60\00\00\60\00\00\60\00\00\60\00\00\60\00"
  "\00\60\00\00\60\00\00\60\00\00\60\00\00\60\00\00"
  "\60\00\00\60\00\00\60\00\00\60\00\00\60\00\00\60"
  "\00\00\60\00\00\60\00\00\60\00\00\60\00\00\60\00"
  "\00\60\00\00\60\00\00\60\00\00\60\00\00\60\00\00"
  "\60\00\00\60\00\00\60\00\00\60\00\00\60\00\00\60"
  "\00\00\60\00\00\60\00\00\60\00\00\60\00\00\60\00"
  "\00\60\00\00\60\00\00\60\00\00\60\00\00\60\00\00"
  "\60\00\00\60\00\00\60\00\00\60\00\00\60\00\00\60"
  "\00\00\60\00\00\60\00\00\60\00\00\60\00\00\60\00"
  "\00\60\00\00\60\00\00\60\00\00\60\00\00\60\00\00"
  "\60\00\00\60\00\00\60\00\00\60\00\00\60\00\00\60"
  "\00\00\60\00\02\7f\7f\03\82\80\80\80\00\01\00\07"
  "\88\80\80\80\00\01\04\6d\61\69\6e\00\00\0a\8f\80"
  "\80\80\00\01\89\80\80\80\00\00\02\c1\00\0f\0b\1a"
  "\1a\0b"
)
(module instance)
(assert_return (invoke "main"))

[titzer]

Dang, complex block types are tricky. I have a partial fix in #289 , but it doesn't handle the case where either ref or ref null are padded LEBs. It fixes this issue though, too.

Thanks for banging on this, we could definitely use a lot more tests in this area.