On Mar 12 09:38, Kaz Kylheku wrote:
> On 11.03.2016 22:27, Marco Atzeri wrote:
> >On 12/03/2016 00:10, David Stacey wrote:
> >>On 11/03/16 14:23, Girish Joglekar wrote:
> >>>I have cleaned up the test example based on X-Windows/Motif which
> >>>crashes
> >>>on 64-bit cygwin cygwin/X but runs on the 32-bit version. The only
> >>>warnings are of the type 'variable set but not used'.
> >>>
> >>>Here is the link to the tar file.
> >>>
> >>>https://www.dropbox.com/s/lknuoac64hfql4s/test.tar?dl=0
> >>>
> >>>Hope you can reproduce the behavior and fix the problem.
> >>
> >>I tried this on Fedora 23 64-bit and it works. I had to add an empty
> >>utcrcu.h file, as the code tries to #include this, and it doesn't exist
> >>in the tar file. Given that it runs in 64-bit Linux, the problem is
> >>unlikely to be an issue associated with porting 32-bit applications to
> >>64-bit.
> >>
> >>Dave.
> >>
> >it can be a bug not triggered on linux due the difference in managing
> >function arguments in the stack.
> 
> Among a myriad other things! Undefined behaviors can have totally different
> consequences even on architecturally similar systems. E.g. a malloced
> buffer overrun depends on the heap structure and allocation pattern
> driven by program inputs. Uninitialized memory will have different garbage
> in it. And so on.
> 
> So it works on Fedora 23 64-bit? That could be by fluke.
> 
> How about this: does it run with zero errors under Valgrind on the same
> system?
> 
> That represents a greater confidence that "it works".
> 
> In any case, this is a poor approach to problem solving in the
> context of open source:
> 
> "Hi, on your platform the program in my multi-megabyte archive crashes
> ... somewhere in the code (I could tell you where, but I can't be bothered
> to use the debugger). Please download the whole, reproduce the problem, and
> debug it for me."
> 
> This kind of burp-and-diaper changing is done in the industry only for
> paying customers.

Heh.  Just for completeness, though, I would like to point out the FAQ I
wrote back when 64 bit Cygwin was still in development:

https://cygwin.com/faq/faq.html#faq.programming.64bitporting

There are a couple of points there which might be interesting.

Additionally we have some experience with actual bugs in the code which
were covered in Linux but showed up on Cygwin.  A prominent example was
a missing declaration of a stdargs function in OpenLDAP which led to
different results on Linux (working) and Cygwin (crashing) because of a
subtil difference in the ABI (6 argument registers in SysV-ABI vs. 4
register arguments in the MS-ABI.


Corinna

-- 
Corinna Vinschen                  Please, send mails regarding Cygwin to
Cygwin Maintainer                 cygwin AT cygwin DOT com
Red Hat