On Tue, 29 Aug 2017 23:46:00 +0200
ludo@gnu.org (Ludovic Courtès) wrote:

> Hello!
> 
> Christopher Baines <mail@cbaines.net> skribis:
> 
> > On Tue, 29 Aug 2017 11:55:08 +0530
> > Arun Isaac <arunisaac@systemreboot.net> wrote:
> >  
> >> Christopher Baines writes:
> >>   
> >> > Modify the install phase to detect when nothing has been
> >> > installed, and error if this happens. This is preferable to
> >> > continuing, and allowing the next phase to fail.
> >> >
> >> > Also, when nothing can be found to be installed, print out each
> >> > file that was considered, along with the regular expressions that
> >> > were used to include and exclude it.
> >> >
> >> > * gnu/build/emacs-build-system.scm (install-file?): Add
> >> > additional error checking and logging.
> >> > ---
> >> >  guix/build/emacs-build-system.scm | 45
> >> > ++++++++++++++++++++++++++++----------- 1 file changed, 32
> >> > insertions(+), 13 deletions(-)    
> >> 
> >> I feel that this adds a lot of complexity (lines of code) to the
> >> emacs-build-system checking for an error that can be quite easily
> >> identified and fixed otherwise.
> >> 
> >> WDYT? Maybe, others can comment on this as well.  
> >
> > In my personal experience, I didn't find this easy to identify and
> > fix. For packaging emacs-minitest, I ended up writing this to pin
> > down why the emacs-build-system wasn't installing the key file.
> >
> > I think validating that something has been installed is really
> > important, as otherwise the later phases fail in a very unclear way.
> >
> > The extra functionality about explaining why each file hasn't been
> > installed is useful for debugging, and I agree that it adds
> > significant complexity.  
> 
> I agree.  I’m guessing you wrote this after spending a while
> debugging a build, despite being experienced with Guix, which to me
> suggests that this is a welcome improvement, in spite of the extra
> complexity.
> 
> > But, I'd like for packaging emacs things to be really easy in the
> > general case, and I think making the build system more helpful when
> > it fails is one way to improve this. I wouldn't like to expect that
> > you'd need to read the implementation of the build system, or add
> > in your own debugging code just to package a emacs module.  
> 
> Sounds reasonable.
> 
> To me this looks like a step in the right direction.

Ok, thanks for your review Ludo :)