Re: CVS commit: pkgsrc

[Aleksej Saushev <asau%inbox.ru@localhost>]

Alistair Crooks <agc%pkgsrc.org@localhost> writes:

> And, having had a chance to sleep on it, some thoughts:
>
> 1. our basic unit of addition/deletion/transfer is the binary package,
> and there's no way we can walk that back. PKGSRC_KEEP_BIN_PKGS will
> therefor stay as it is
> 2. not sure how many people are using pkgsrc read-only, but it's a
> useful thing and keeps us honest, so we should be supporting it

Thus, 2016Q3 branch is broken and we have to implement some workaround.
How nice...

> 3. if you want to bootstrap, you'll want the binary packages that are
> made from it. So PACKAGES should be set before bootstrapping.
> Interesting to note the "make binary kits" arguments to bootstrap.

No, you probably won't, actually. Binary bootstrap kit is the main result,
and it is enough.

> 4. I think it's a bug that PACKAGES can't be set as an argument when
> calling bootstrap, so diffs attached which do this
> 5. tested on NetBSD only (all I have ready access to right now), and
> not tested with a read-only pkgsrc. However, it DTRT when
> bootstrapping with defaults and when explicitly setting via --packages
> on the command line (including creating the ${PACKAGES} dir itself)
>
> I'd prefer to wait until after 2016Q3 is branched before applying
> these, as it is likely to cause infrastructure packages to be rebuilt
> if applied now - easier to wait until after the branch, and then pull
> up to branch when tested by everyone.

...

> @@ -1056,6 +1063,7 @@
>  fi
>  
>  # save environment in example mk.conf
> +echo "PACKAGES=                      $packages" >> ${TARGET_MKCONF}
>  echo "PKG_DBDIR=             $pkgdbdir" >> ${TARGET_MKCONF}
>  echo "LOCALBASE=             $prefix" >> ${TARGET_MKCONF}
>  echo "VARBASE=               $varbase" >> ${TARGET_MKCONF}

That's definitely a significant regression. It kills possibility of passing
PACKAGES via environment, for instance.


-- 
HE CE3OH...