Support for "partial upgrades"

Mikhail T. mi+thun at aldan.algebra.com
Mon Jan 5 22:47:02 UTC 2015


Hello!

A prominent committer stated today
<https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=196518#c5>:

    "We do not support partial upgrades, never had, never will."

He then elaborated, that by "partial upgrades" he meant building a port from an
updated tree without first rebuilding all of the already-installed ports.

Whatever the deal may be with perl5-versions (subject of the above PR), I'd like
voice my concern about the "do not, never had, never will" part. For we
certainly "had" supported such things. For example, the ability of a port to
LIB_DEPEND on a shared-library without a specific major-number
<https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=34126> was added -- 12 years
ago -- exactly to better support such a "partial upgrades":

 1. Build everything.
 2. Update your ports-tree.
 3. Try to build one more thing.

And we still "do" support such handling of LIB_DEPENDS -- even if another
prominent committer does not know about it
<https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=196519#c1>.

While no "guarantees" can be provided by a free software project (indeed, even
commercial ones usually do not), it is rather unreasonable to demand, the user
rebuilds /everything/ in order to be able to build one more port from an updated
tree...

I'd argue, that one simply can not comply with this policy, while maintaining a
FreeBSD system usable for anything /other/ than testing the rebuilds themselves
and show-casing... Unless, of course, portmgr@ wants us all to switch to
prebuilt binaries (and Koolbuntu), that is...

It is one thing to say "yeah, this might not work -- send us patches": we may
not be able to /afford/ the ideal (due to shortage of people and/or material
resources), but we agree on /what the ideal is/. It becomes different, when the
response is "this should not work, never worked, never will"...

I'd like to see us continuing to recognize, that one FreeBSD install may differ
from another, and ports not having gratuitously-strict dependencies and
requirements... Could portmgr@, please, (re)affirm this goal and otherwise
clarify the matter for the benefit of mortals and the body's own members alike?
Thank you!

    -mi



More information about the freebsd-ports mailing list