Re: [Caml-list] Some comments on recent discussions

[Gerd Stolpmann <info@gerd-stolpmann.de>]

Am Freitag, den 16.12.2011, 15:11 +0100 schrieb Alain Frisch:
> On 12/16/2011 02:14 PM, Gerd Stolpmann wrote:
> > Simple answer: There is a bootstrap problem: The existing Ocaml users
> > are almost Unix-only. They do not care about Windows. In order to
> > establish "Windows-typical problem solving" you need definitely more
> > Windows users, but they will only come if you have a Windows-typical way
> > of distribution.
> >
> > My thinking is that you can break this circle only if you go forward and
> > try to make as many Unix-style solutions available under Windows as
> > possible.
> 
> Honestly, I think this is the wrong approach.  The Windows-style of 
> packaging (i.e. lack of packaging system) is very simple to implement: 
> just ship zip files or Windows installers. It's less comfortable than 
> powerful package systems as we're used to in the Unix world, but it is 
> not an issue for Windows users.
> 
> > By the way, your plan includes Unix emulation, too, under point 3. It's
> > only more hidden.
> 
> Yes, but Unix emulation exists and works not too badly. You can use 
> Cygwin, or msys, etc.  I can see two reasons for reimplementing 
> Unix-like tools, and none of them are good:
> 
>   - We want to control everything in order to have better options for
>     creating nice packages for OCaml related stuff. This is a bad reason,
>     because Windows users except binary package anyway, so the external
>     tools needed during the build process don't have to be redistributed.
>     The only external tools which are currently mandatory to use ocaml
>     are an assembler and a linker, and some native libraries. Not
>     a Unix shell, a "find" or a "ls" command.
> 
>   - Existing solutions are not build on strong technical foundations,
>     and one thinks one can do better. Ok, but this has nothing to do
>     with OCaml; this will necessarily represent a huge amount of
>     work; and I don't see why people who don't really care about Windows
>     would do a better job in a few months than people who've spent years
>     of work on cygwin/msys/etc.
> 
> 
> Can you explain what's wrong with my approach:
> 
>    * OCaml users have access to binary packages which can be trivially
>      installed (zip files or installers) and are self-contained.

I have no problem with that.

>    * People in charge of creating these packages use existing Unix-like
>      environment (cygwin/msys/...); and progressively, they
>      publish "best practices" for the whole community in order to
>      simplify their job (e.g. try to rely on ocamlbuild or omake instead
>      of Makefiles; avoid relying on symbolic links; etc).

That's the part where I doubt it will happen. How do you find such
people? The job description does not sound very attractive, and
honestly, if I don't have any (or only limited) interest on my own in
this, I'd do this type of work only for money (because it's frustrating,
and Windows users are not really known to be grateful for getting
something for free).

Hence, my idea of reusing as much Unix scripting as possible. It's
finally easier.

Btw, a lot of infrastructure already exists: GODI for mingw, only that
it is broken now, and would need a few days of fixing everything.
Because GODI includes a binary package concept, you could easily extract
the built libraries, and distribute them in a zip-style format.
Currently you need the mingw toolchain for using the compilers, but as
you pointed out, this could be improved.

You may ask why GODI is broken for mingw. Well, see above. The job is
unattractive, too time-consuming, and I lost all interest in it. It's
better done by somebody who needs that for his/her own work.

Gerd

> 
> 
> 
> Alain
>