Re: [Valgrind-developers] PATCH: Make Calltree work without --pointercheck=no (2)

[Julian S. <js...@ac...>]

> Can somebody (Jeremy?) review it and commit it if it is OK?

Just discovered "Kompare".  Nice tool!

At least in the details, it looks OK.

I am a somewhat concerned about the type system abuse that's
already there, though (nothing to do with your patch).

   static void emit_movb_reg_regmem ( Bool bounds, Int reg1, Int reg2 )

First arg is Bool, conceptually 1-bit and in fact unsigned char
(see include/vg_skin.h).  But then there are calls

   emit_movb_reg_regmem ( size & DO_BOUNDSCHECK, reg1, reg2 ); 

This just can't be right, and I'm not even sure what the caller
intends here.  Not to mention DO_BOUNDSCHECK is 1 << 8 so is 
zero when interpreted as Bool.  If the bounds check always happens
for instructions emitted by this call, it's because size just 
happens to always be 1, 2 or 4 but never zero -- and completely
nothing to do with DO_BOUNDSCHECK.

I guess the C "type system" (so-called) gets to wear the pointy 
hat and stand in the corner :-)  Anybody know of a way to define
a Bool type in C in such a way the compiler can't silently confuse
it with any flavour of int?

J