Jeremy C. Reed wrote:

> On Sat, 18 Jan 2003, Amitai Schlair wrote:
> 
> > Likewise, I've avoided bumping PKGREVISION when updating a package
> > to build and install on Darwin. As I see it, PKGREVISION is a way
> > to tell the user (or pkgsrc itself) that a previously working
> > package has been improved. If the package just stopped being broken,
> > the user (and pkgsrc) already know it's been improved.
> 
> I believe it should be increased then too.
> 
> I guess it depends on definition of "broken".
> 
> A "broken" package may have been built (packaged), but just not work
> properly.
> 
> So the version number needs to be different so an end-user easily knows
> which package is which.

Yes, it depends on in what manner the package had been broken. The
scenario in which I don't bump PKGREVISION:

Before:
* NetBSD: package works
* Darwin: package doesn't build and/or install

After:
* NetBSD: package works exactly as before (no user-visible changes)
* Darwin: package builds and installs properly
* Other platforms: no change

In this scenario, the only user-visible change in the package is
that on Darwin, it can be installed, whereas it couldn't before.
So if it doesn't work on Darwin, try cvs updating. Then, if it
works, you know that 1) it's been repaired and 2) you have the
"revision" with the repairs included.

To put it another way, if you couldn't even register the package
before, there's no PKGREVISION to compare.

It doesn't hurt much to bump PKGREVISION in this case, but I
don't see how it would help. I'm always ready to be persuaded,
though!

- Amitai