Re: gnustep-make experiment

[Richard Frith-Macdonald]

On 13 Feb 2007, at 13:39, Andrew Ruder wrote:

> On Tue, Feb 13, 2007 at 06:18:54AM +0000, Richard Frith-Macdonald  
> wrote:
> > An extra dependency most emphatically IS an issue ...  because the
> > 'people' you are referring to actually just means 'you', and you are
> > just guessing about other users, and even assuming that 'most' is
> > actually the case, then what you are proposing is to have something
> > that only works for 'most', not for all.  Is your guess that  
> > 'most' is
> > 55%, 70%, 95%?  At what point is the  remaining 45%, 30%, 5%
> > sufficiently small a minority that they become  a non-issue?
>
> First off, pkg-config is a developer dependency.  Users of gnustep
> applications that have them packaged by their distribution will not  
> have
> to worry about pkg-config.  For what it is worth, some of our
> dependencies are already using pkg-config (libxml2, sqlite3, libart,
> libpng, freetype, etc.).

Is the implication here that developers using a cross platform  
development package don't matter?

> Second off, as pkg-config is a *developer dependency* far less worry
> should be put into making it a dependency.  If you are an actual
> developer, I'd hope you have the capability to get a few extra
> dependencies sorted (especially considering that pkg-config IS  
> available
> on so many distributions at this point).

Sure I can, but I shouldn't have to unless I gain some considerable  
benefit from doing so.

> > GNUstep is supposed to be a cross-platform development system, and
> > anyone who actually does cross platform development would hit a
> > problem with pkg-config.
>
> Are we talking about windows here or what?  Seems like most of the  
> time
> cross-platform means windows,

Windows, Solaris, MacOS-X that I know of.
I guess most gnu/linux systems have it, but most others don't.

> and once again, pkg-config should be able
> to be compiled/setup on windows.

That's missing the point.  Unless we supply it with the system and  
automatically install it, it's another hurdle.  We already have too  
much installation of dependencies to do (particularly on windows),  
and should be working to automate things to make existing  
dependencies less painful.

> > Dependency issues do not rule out using things (indeed, GNUstep has
> > plenty of dependencies), but they are a big factor to consider, and
> > only appear minor to the people they don't happen to effect, so when
> > adding a dependency we need to be clear that what we are adding has
> > overwhelming advantages over the alternatives.  While pkg-config is
> > quite nice (and looking at it provides inspiration), it has no such
> > advantages.
>
> There is one major advantage that pkg-config does provide here and  
> that
> is simply that we are not maintaining it.  Everytime we work around
> having another standard system service by reimplementing it in our
> codebase, sure we save a 'dependency', but at the same time we have  
> more
> code that we have to support and that is never a good thing.

Well, that's a different issue than dependencies, but as Nicola  
pointed out, using pkg-config means we have to support it in terms of  
providing the metadata files it depends upon ... and that's actually  
as much work as (or more than) just building our own stuff.  So in  
terms of maintenance and development effort required, the use of pkg- 
config is an additional burdon, not a time saver.