RE : RE : [Pnet-developers] [bug #4430] Internal error: Could not verify the code - br.s at verify_branch.c:393

[Yannis BRES]

     Hi again !
     Here is (I guess) interesting feedback concerning OUR bug, stated as
"Non-compatible types on stack depending on path" by peverify.
     So, very roughly, we had a branch of a conditional that would exit (but
the bytecode verifiers do not known that), leave on the stack the object
returned by our exit function, and join another branch that leaves an
integer on the top of the stack.  Don't ask me why this is the only bytecode
compliance error in a >50MB IL file...
     So, both ilverify and peverify do complain about that.  However, upon
executing the invalid program, only the PNet on-the-fly bytecode verifier
complains, the MS one do not.  I guess that this is because the branch that
exits is never taken for the input file I used.  I'm surprised that MS
bytecode verifier do not entirely check a method upon calling it, and I
guess that they do not completely JIT it as well.  Maybe they only work at
the basic block level, which seems interesting...
     That was just food for thought ;-)
          Yannis


======================================================================
ESSI                                    http://www.essi.fr
Attache Temporaire d'Enseignement et de Recherche
----------------------------------------------------------------------
INRIA  -  Equipe Mimosa                 http://www-sop.inria.fr/mimosa
Collaborateur Exterieur                 +33 (4) 92 38 79 67
----------------------------------------------------------------------
+33 (6) 60 38 79 52                     http://www.yannis.bres.name
======================================================================