Re: stublibs and bytecode only packages. Was: Re: Plans [Re: Cameleon 1.0]
On Sat, Sep 07, 2002 at 11:33:22AM +0200, Stefano Zacchiroli wrote:
> On Fri, Sep 06, 2002 at 07:15:25PM +0200, Sven LUTHER wrote:
> > > Am I missing something?
>
> < snipped a lot of useful information >
>
> Ok, thanks for the hints.
>
> Anyway I was reasoning mainly about libraries (even if I forget to
> mention it, sorry).
> Libraries which ships .so can't be arch: all, but have to be
> architecture dependent, this was may point.
Yes, but the reason to build the .so is so programs can be made arch:
all, something upstream is always afraid to do, since they have not the
control we have on the libraries, they prefer shipping the programs with
all the stuff included.
> On the other hand, we can surely state in the ocaml packaging policy to
> build, where possible and useful (i.e. not cpu bound programs),
> architecture independent ocaml based _programs_, this will be surely an
> improvement for the user and for the spreading of ocaml programs.
Yes, and even go beyond that, every ocaml program should be built as a
split binary package, where the package is built in bytecode, and a
-native or something version is also built on the archs supporting
native code compilation, and would divert the executable from the
bytecode version or something.
> Regarding the additional .debs that the user have to download (mainly
> the additional 'ocaml-base' package, IMO this is not a problem because
> it have to be downloaded only once and promote future reusability.
Yes, but it may also be lablgtk, i think.
> I'm really in favour of trading off one-package download overhead with
> code sharing between many packages, this is also the philosophy behind
> system shared objects.
Yes, as you may have seen, i have already uploaded a arch: all ledit
package (1.10-6).
Friendly,
Sven Luther
Reply to: