Re: [Tinycc-devel] tinycc fails to build mksh because of bound check

[Thomas Preud'homme]

Le vendredi 27 août 2010 16:28:03, grischka a écrit :
> Thomas Preud'homme wrote:
> > But finally, I found it was related to bound check code.
> > Indeed, when disabling the code the compilation works like a charm. There
> > was a lot of bound_add_ptr and other functions related to bound check. I
> > guess one of the function return a bad result or simply do something
> > wrong by accessing a zone out of bound (which would be quite paradoxal).
> 
> Just before you spend too much time on it, the algorithm itself used
> by the bound checker has been said to be not 100% reliable.  I don't
> know why and where though.  My impression is that it's okay for the
> few test cases given in boundtest.c, however not ready for compilation
> of real programs (probably not even tcc itself).
Note that the program is not compiled with -b. It fails only because 
CONFIG_TCC_BCHECK is defined (which it is by default on i386). So if there is 
a problem with that, CONFIG_TCC_BCHECK should not be defined for any 
architecture by default.
> 
> > ... and if it's quick to fix I'll also take a look to the function
> > array problem [3].
> 
> I tested on windows and the pointer is set there which could mean there
> is something specific going on with ELF symbols or relocation.
So that's definitely a bug for me :)
> 
> --- grischka

Thomas Preud'homme

signature.asc
Description: This is a digitally signed message part.