Re: pkg build: PKG_ADD run?

[Olaf Till]

On Sat, Mar 16, 2013 at 11:03:04PM +0000, Carnë Draug wrote:
> On 16 March 2013 22:34, Olaf Till <address@hidden> wrote:
> > On Fri, Mar 15, 2013 at 04:54:16PM +0000, Carnë Draug wrote:
> >> On 15 March 2013 09:38, Olaf Till <address@hidden> wrote:
> >> > Since PKG_ADD is meant to be run when the package is added to the
> >> > path, I'd consider it a bug if it is run even if only a binary package
> >> > is built. Do you see any reasons to run PKG_ADD even in this case?
> >>
> >> I'm not 100% of how PKG_ADD is ran when doing `pkg build` but I'd
> >> guess it may make sense for pkg to try to install and load the package
> >> it just built, to make sure it's built correctly.
> >
> > I see now that the problem with the package was triggered by the fact
> > that a PKG_ADD instruction of an oct-function required both the
> > oct-function and the packages m-functions to be in the path. Since in
> > 'pkg install' the m-functions seemingly are (happen to be?) added to
> > the path before the oct-functions (adding to path triggers respective
> > PKG_ADD instructions) this is no problem here. But in 'pkg build',
> > since the functions are not already in the path due to installation,
> > 'gen_doc_cache' will temporarily add them (triggering PKG_ADD
> > instructions), but m-functions and oct-functions separately, so they
> > are not in the path at the same time; so there is a problem.
> >
> > Should this be fixed in Octave? If yes, the only way I see is to make
> > 'gen_doc_cache' return each directory it newly added to the path
> > (instead of removing it from the path again), so that all directories
> > of a package temporarily added to the path can be removed from the
> > path at the same time by the caller in pkg.m, after all gen_doc_cache
> > calls are over. Should I make such a patch?
> >
> > The problem could also be worked around in the package by providing
> > m-file wrappers for the oct-functions.
> 
> Yes there is a problem. I remember it from the io package which had
> the opposite issue. Since the path for oct was loaded first, and the
> PKG_ADD had a function from the inst directory it would fail. The work
> around we still have on the IO package is to have a post_install
> script https://sourceforge.net/p/octave/code/9983 removing the
> unnecessary PKG_ADD file.

I'll probably make a workaraound anyway, but there could be also the
fix in Octave to, a.o., modify addpath. addpath already accepts
multiple directories as arguments; it could probably be changed to
trigger PKG_ADD instructions only after adding all directories to the
path.

Olaf

-- 
public key id EAFE0591, e.g. on x-hkp://pool.sks-keyservers.net

signature.asc
Description: Digital signature