Re: [Gnu-arch-users] Features command for arch

[Robert Anderson]

Matthieu Moy wrote:

> "Robert Anderson" <address@hidden> writes:
>
>  
>
> > is being used from last time 'configure' was run.  If you're
> > going to do something really crazy like use a different tla every
> > time you run the front end, the front-end could hash the
> > executables to an array of configurations.
> >    
>
> That means each front-end has to implement a persistant table (hash ->
> features) or at least a persistant checksum.
No.  That can go in the configure process itself (along with any such 
scheme of anyone's choosing).

> One more config file, one more dependancy on md5sum, ...
Look, the configure information has to be somewhere.  It seems pretty 
unconvincing
to complain about "another config file" when the alternative is some 
data structure in tla.

> Developing xtla, we currently face this kind of problems, not for tla,
> but for the differences between Emacs versions. And believe me, the
> ability to "introspect" your own features in an extension language
> (featurep, fboundp in the case of Emacs lisp) is _much_ easier to
> implement than a configure script !
emacs is fundamentally different in that it primarily interacts with a 
display.  tla primarily
interacts with the shell and the file system.  _Big_ difference in how 
well it interacts with a configure
script.

> One day, we'll have to manage the different versions of tla, and I'd
> clearly prefer testing tla's features with a standard way than typing
> a new testcase myself in a configure script to get the answer.
Well, of course.  That offloads the burden on somebody else.  I just 
don't think that's where the
burden belongs.

Bob