Re: [Chicken-users] Subversion support misguided IMHO

[felix winkelmann]

On 1/6/07, John Cowan <address@hidden> wrote:
> Brandon J. Van Every scripsit:
>
> > Apparently, GHC builds are dicey.
>
> It's GHC *ports* that are hard.  GHC is a full-bore compiler, not a
> C generator like Chicken, so if you are trying to port to a new
> architecture you wind up writing a new set of code-generator tables.
> You can also compile to C, but there are compromises.

Well, I could go to great lengths about that, but I don't want to start
ranting...

> > The vast majority of builds out there, totally suck.
>
> My experience is quite otherwise: the vast majority of Unix builds
> work fine on Unix, and most of them even work fine on Cygwin as well.

Sorry, but I have to disagree. I work in a company that does support
for projects
that use Linux in embedded systems. Mostly this revolves about building kernels,
applications and complete customer- and application-specific
distributions. This
means that 80% of the time we are building UNIX software, porting compilers
and libraries to new platforms and build toolchains.

It is a complete, horrible, incredible mess! Even without
cross-compilation, many
(well known) software packages break easily, due to various reasons (autotools
versionitis, compiler quirks, broken Makefiles, etc.). Add cross-compilation and
things get even worse. Did you know that OpenSSL can not be properly cross-
compiled? Sure, you can hack around it, as one can nearly always in a UNIX
environment, but the maintainers still refuse to fix it.

And that's only on Linux. Yesterday I built FLTK on my Mac. I had to
fix two bugs
in the Makefile before I could get shared libraries. Today aalib (also on Mac).
I had to pass an obscure hostspec on to configure (not the default one!) and
remove #includes of "malloc.h" before it worked. These are all small
things, not really critical, but that's quite different from "working fine".

The UNIX bonus is that with some knowledge and time one can work
around all of these
problems, and that is the only reason why people are not simply giving up (and
because this knowledge is freely available, as opposed to, say, Windows).

Yet I hope that this is not the optimum. There must be a better way. I don't
know which one, but I can't accept that what we currently have is the best
that's available. A VM perhaps, but look at Java - it's just the same mess in
a different disguise.


cheers,
felix