Le 16 août 07 à 13:47, Rainer Müller a écrit :
N_Ox wrote:
Le 14 août 07 à 19:58, N_Ox a écrit :
Le 12 août 07 à 05:28, N_Ox a écrit :
From what I've understood, the problem with default variants is that if we do something like `port foo -bar` and that foo has `default_variants +bar`, if an upgrade of foo is released, then `port upgrade foo` will enable bar variant because variant disabling are not saved and `default_variants +bar` will be evaluated.
Is this really a problem? Can't we use the magic of the TCL interpreter and ignore the call to default_variants procedure in upgrade variant?
These were my $0.02 for tonight... ...I'll now take my pills and go to bed :p
Another problem with default_variants are variant conflicts. Something among those lines:
variant apache2 {...} variant apache conflicts apache2 {...} default_variants +apache2
and this port command: port install +apache
would produce an error because that would select apache and apache2.
So here is my idea: let's create a new block named variant_group, or variant_choice, or even both.
variant_(group|choice) webserver { variant apache2 {...} variant apache {...} }
variant_group would tell us that we _can_ enable support for _one_ webserver, namely apache or apache2. variant_choice would tell us that we _must_ enable support for _one_ webserver, namely apache or apache2.
That's all for today. Regards,
Almost all... I've forgotten something:
default_variants +apache2 and port install +apache would not fail and would automagically select apache instead of +apache2.
No one commented on this yet, so I will do start the discussion. I think this is a very reasonable idea to group variants to make it easier to implement choices. I also remember that we have some port having a choice between gnutls and openssl, so this is really a common use case. But there is still the need to record disabled default variants in order for this to work.
Rainer
As I've said in the very first mail of this discussion, I don't think recording disabled default variants is really a problem. The only case it is one is during upgrade, and what I've said should fix that. -- Anthony Ramine, the infamous MacPorts Trac slave. nox@macports.org