Variants handling: my $0.02.

[Jordan K. Hubbard]

On Feb 4, 2008, at 11:23 AM, N_Ox wrote:

> [ ... description of new variant syntax ...]

First, let me just say how happy I am to see people revisiting the  
topic of variants and how to better express them.  I think variants  
are one of the more powerful features of MacPorts (that no other port  
systems, to my knowledge, share), but also the most likely to cause  
confusion and mayhem if misapplied across the collection (I think  
Landon went through periods of both pride and remorse over variants  
for just this reason).

Second, given that the "platform" procedure also feeds directly into  
variants, I hope that cleaning up variants will include cleaning up  
the platform syntax and general mechanism.  More specifically, as I  
pointed out a couple of weeks ago, we're starting to see a lot of  
duplicated code across platform clauses because it's not possible to  
have wildcards in platform clauses or even just be able to do platform  
includes, e.g.:

platform darwin 8 i386 {
	platform darwin 7;	# include the generic platform code for darwin 7
	... code specific to darwin 8/i386 here ...
}

Given how frequently the platform clauses are being used, and the fact  
that they're only going to get more prolific as further releases of  
MacOSX come out, I'd say cleaning them up is at least as important as  
cleaning up variants (which, again, they just use at the back-end  
anyway).

- Jordan


>
> Le 4 févr. 08 à 18:13, Rainer Müller a écrit :
>
>> N_Ox wrote:
>>> These two sugar syntaxes would make the variant writing process  
>>> cleaner.
>>> But maybe they could help us more...
>>> Let's say the variants which do something only when they are  
>>> disabled (variant -some_variant) are always enabled by default.
>>> In this setup, `sudo port install some_port +another_variant`  
>>> would install some_port @some_version+another_variant+some_variant.
>>> In the registry, we would save "some_variant another_variant" as  
>>> the list of selected variants.
>>> Now, let's explicitely disable the variant: sudo port install  
>>> some_port -some_variant +another_variant.
>>> In the registry, we would save "-some_variant another_variant".
>>
>> Should the selection of -some_variant be user visible or just  
>> stored internal?
>> port installed some_port
>> 1) some_port at 1.0+another_variant-some_variant
>> 2) some_port at 1.0+another_variant
>
> Shown to the user
>
>> The main problem about default_variants in it's current  
>> implementation is that we don't store disabled variants and re- 
>> select them on upgrade...
>
> AFAIK, default_variants is just a procedure that append any argument  
> to the list of enabled variants in the registry.
> This syntax remove the need to use default_variants.
>
>>> No need to use default_variants.
>>
>> So how would I tell that a +variant is default? E.g. the lynx port  
>> provides support for OpenSSL (+ssl) and GNU TLS (+gnutls). They are  
>> conflicting variants, with +ssl in default_variants. But the user  
>> has the choice to install it with GNU TLS by using -ssl +gnutls.
>>
>> But how would I write that in the new syntax?
>>
>>> No more "Why the hell this variant is enabled when I upgrade this  
>>> port?" whining.
>>> No more "no_x11" variant:
>>> variant -x11 {
>>>    # no x11
>>> } else {
>>>    # x11
>>> }
>>
>> And which of those will be the default...? You said -some_variant  
>> will be enabled by default, or did I understand that wrong? In  
>> which cases will -some_variant be enabled by defaul? If it got no  
>> else-block?
>> I think that will end up more complex than it currently is...
>
> Not complex, at least that's not what I think.
> The algorithm would be:
>
> foreach {name} ${variants} {
> 	# Foreach variant defined in the portfile
>
> 	if {[lindex ${selected_variants} ${name}] > -1} {
> 		# If this variant is selected.
> 		eval_variant ${name} # The first block.
> 	} elseif {[variant_has_else_block ${name}]} {
> 		eval_else_variant ${name} # The else block
> 	}
> }
>
> If a -variant hasn't been SELECTED (through the command line), it is  
> ENABLED, it did NOT show up in `port installed` and the else block  
> will be evaluated.
> If a (+)variant hasn't been SELECTED (through the command line), it  
> is DISABLED, it did NOT show up in `port installed` and the else  
> block will be evaluated.
>
> So the behaviour is in fact the very same for both of them.
>
>
>> Also, what is the difference between these and which one is a  
>> default variant?
>>
>> variant foo {
>>   # do something
>> }
>
> A normal variant:
> port install		-> @...		(nothing evaluated)
> port install +foo	-> @...+foo	(first block evaluated)
> port install -foo	-> @...		(nothing evaluated)
>
>> variant -bar {
>>   # do something
>> } else {
>>   # do something
>> }
>
> A negative variant which does something else when it's not enabled:
> port install		-> @...		(else block evaluated)
> port install +bar	-> @...		(else block evaluated)
> port install -bar	-> @...-bar	(first block evaluated)
>
>> variant -baz {
>>   # do something
>> }
>
> A negative variant:
> port install		-> @...		(nothing evaluated)
> port install +baz	-> @...		(nothing evaluated)
> port install -baz	-> @...-baz	(first block evaluated)
>
>> What if we add some new flag? I like the idea of variant - 
>> some_variant, but maybe we could also add a default keyword which  
>> says if this variant is going to be installed by default.
>>
>> Like this:
>> variant x11 description {Install X11 support} default {
>>   # x11
>> } else {
>>   # no x11
>> }
>
> Sure. The -variant syntax just sounds more intuitive to me, but one  
> could say I'm strange.
>
>> I like the overall idea to simplify the default variants handling!
>>
>> Rainer
>
> --
> Anthony Ramine, the "Ports tree cleaning Maestro".
> <nox at macports.org>
>
> _______________________________________________
> macports-dev mailing list
> macports-dev at lists.macosforge.org
> http://lists.macosforge.org/mailman/listinfo/macports-dev