Re: pkg / PKG_ADD / arch-dependent

[Philip Nienhuis]

Olaf Till wrote:
> On Sat, Mar 08, 2014 at 04:44:29AM -0800, Philip Nienhuis wrote:
> > Julien Bect wrote
> > > Hi everyone,
> > >
> > > Is there any reason why the PKG_ADD/PKG_DEL scripts located at the root
> > > of a package source (.tar.gz) should end up being moved to the
> > > arch-dependent subdirectory after installation with "pkg install" ?
> > >
> > > Because this is the behaviour that I currently observe while trying to
> > > create my first octave package (Octave 3.8.1 / Ubutu 13.04).
> > >
> > > The documentation here :
> > >
> > > http://www.gnu.org/software/octave/doc/interpreter/Creating-Packages.html
> > >
> > > doesn't say explicitely where PKG_ADD will be, just that it is "run when
> > > the package is added to the users path".
> >
> > If there is a arch-dependent subdir (some OF packages don't have one),
> > PKG_ADD will end up there. Otherwise PKG_ADD ends up in the scripts subdir
> > (AFAIK, I'm not sure).
> >
> > Yep, sometimes this behavior is very undesired.
> > E.g. in the io package, PKG_ADD (in the arch-dependent subdir) was already
> > run when the package hadn't fully been loaded yet, so a number of commands
> > needed for proper operation of the io package couldn't be executed.
> > That's why in post_install.m I entered some stanzas that move PKG_ADD to the
> > scripts subdir.
> >
> > If it solves your problem you can copy those lines to your package's
> > post_install.m.
> >
> > A better solution would be some flag in e.g. the file DESCRIPTION in the
> > package .tar.gz telling pkg.m where PKG_ADD is to be moved to.
>
> AFAIK the arch-debendent path is added (with addpath) after the
> scripts path in 'pkg install', so since the rootdir/PKGADD directives
> get into the arch-dependent directory, they won't be executed until
> both the scripts directory and the arch-dependent directory are in
> Octaves path.
> This seems reasonable to me and shoudn't cause problems
> in 'pkg install'.

It's not so much during 'pkg install', but rather during "regular" 
loading of a package (e.g., packages that are auto-loaded in the course 
of starting Octave) that problems may occur.

It could be that "regular" loading of a package shouldn't differ from 
what happens during 'pkg install'; but my experiences (and Nitzan's, who 
made the first workarounds for the io package) are that PKG_ADD (and 
PKG_DEL) in the arch-dependent subdir did hamper proper set-up of the io 
package.
Thinking it was a mere timing issue I introduced delays in PKG_ADD 
("sleep (0.5)") but that didn't help. (FYI, the io package doesn't 
invoke any compiled modules during loading, only .m-files.)

But I just checked: moving io's PKG_ADD to the arch-dependent subdir (in 
Octave-3.8.0) doesn't make a difference now, the io package is set up 
fine. Octave's search path neatly has the script subdirs below (= 
before) the corresponding arch-dependent subdirs as you say.

But.... moving io's PKG_DEL into the arch-dependent subdir too does work 
out adversely: that hampers proper unloading of the io package.
It seems the script subdir, containing .m-files required by PKG_DEL 
during unloading, is removed from Octave's search path before PKG_DEL is 
run.
I'd expect that subdirectories should be removed from the path in 
reverse order. Is that the case in pkg.m and friends?

Philip