On 3/31/06, Joel VanderWerf <vjoel / path.berkeley.edu> wrote:
> Austin Ziegler wrote:
> > On 3/18/06, Kaspar Schiess <eule / space.ch> wrote:
> >> Hello Glenn,
> >>
> >>> Question is, which is the best (free) compiler for the job?  Visual
> >>> C++ or Borland? Or another?
> >> I won't say mine's best, however all RMagick binary releases are compiled
> >> using mingw and msys. I think they integrate posix/unix stuff nicely with
> >> windows while guaranteeing binary compatibility with other windows builds.
> >>
> >> Just thought I'd complete the thread with what works for me.
> >
> > Note that mingw/msys builds are NOT compatible with VC++2003 or
> > VC++2005 builds because of a different runtime environment. At a
> > minimum, there is a mismatch between where errno is defined (somehow).
> >
> > If you are using the One-Click Installer for 1.8.2, you will need to
> > compile extensions with VC++2003 (using MSVCR71.DLL); if you are using
> > the One-Click Installer for 1.8.4, you will be able to compile
> > extensions with VC6 or msys/mingw. You MUST NOT mix the versions or
> > bad things may happen to your Ruby programs.
>
> I've done a little work with an extension compiled with VC++2003 and
> ruby-1.8.4, One-Click. Am I just lucky that it works? Or is that known
> to be OK?

You're lucky.

It all depends on what Windows APIs it uses in combination with other
Windows API calls outside of the extension. One known culprit is the
global variable ERRNO which is physically part of the runtime DLL --
and there is a different runtime DLL for each compiler version. So if
you extension calls an API that sets ERRNO, and another part of Ruby
checks the value of ERRNO, they will be using different memory
locations, with unpredictable results.

Curt