Re: [gentoo-dev] Secure Gentoo - OO-ebuilds

[Karl Trygve Kalleberg]

On Thu, 10 Jan 2002 08:58:14 -0500
John Stalker <stalker@××××××××××××××.EDU> wrote:

> Damon M. Conway:
> Since eclasses are associated with a learning curve, would it not be
> preferrable to recast the whole inheritance thing in a proper
> object-oriented language and rather build a support framework for it
> there?
>
> Karl Trygve Kalleberg:
> hmm using scheme sounds very appealing, but the problem is that
> many of us have an unnatural fear of parenthesises... By using
> OO portage would be indeed more powerful and it would make it possible
> for a user to override doman if he wishes so easily, or to add a hook
> procedure that is run whenever src_compile is ready...

You got the names reversed :p

> Me:
> I think we'd better keep portage in bash (or rather what appears
> to be bash).  We also need to keep it fairly simple.  Otherwise we
> lose two important resources: (1) users contributing ebuilds rather
> than just installing tarballs on their own systems and (2) users
> making a legitimate attempt to diagnose bugs in an ebuild rather
> than just filing bug reports like "emerge failed."  I don't see
> a problem in adding hooks.  One can have emerge source config
> files, if present, at various points AFTER sourcing the ebuild
> but BEFORE calling functions and then call before_compile,
> after_compile, etc.  The advantage of something like that is that
> it is pretty much invisible.  If you want you use the hooks,
> override some defaults, etc. and you don't need to bother the
> ebuild writer or users who are happy to accept the defaults.

What you correctly imply is that the targeted user-base for Gentoo can be
assumed to have a fair knowledge of bash scripting. I really don't think
we should assume knowledge of other languages (especially since the second
language for most power users/system administrators seem to be Perl, which
some of us really don't fancy ;).

I am actually fairly sceptical to bringing too much OO thinking and
terminology into our ebuilds. I think that we will for the most part see
that we will have trivially short inheritance chains, and I cannot
off-hand see any use for the other fancy features of OO (polymorphism,
data hiding).

I'd like to have Dan Armak and Bart Verwilst (and any other people with
experience with our eclasses system) to write some sort of summary/report
and post to the list on where eclasses have been found to be beneficial
and whether or not their drawbacks (obscurity, mainly) are too bad.


Extending portage with (possibly context-aware) defaults to src_compile
and src_install will most likely happen at a later stage. If there seems
to be a need for before_compile and after_compile, that might also be
done, but this is currently handled by putting the necessary commands into
the src_compile-function itself.

There has been mentions of a separate src_configure step, though. I think
Mikael knows more about this.


Kind regards,

Karl T