Re: Can we stop revbumping the world please?

[David Holland <dholland-pkgtech%netbsd.org@localhost>]

On Sat, Nov 09, 2024 at 07:22:45PM +0100, Thomas Klausner wrote:
 > Not to disturb a proper rant by boring facts but
 > 
 > fossil -> sqlite3 (optional, which you just removed) -> icu (optional)
 > objfw -> clang (on Darwin) -> libxml2 -> icu (optional)

It's not clear that we need to revbump for non-default options. The
usual rule for bumping the version for direct changes is "is the
binary package different?" For non-default options, the binary package
is not different no matter what you do upstream.

It's not quite that simple because we also use the version bump as
implicit guidance for when you actually need to recompile (as opposed
to the unsafe-depends state, which is severely overconservative in
practice) and if you do turn the options on, that rebuild is needed.
But we ought to do this signalling some other way.

We do a lot of belt-suspenders-maglev-and-antigrav checks and it's
expensive.

Also, quite apart from that, there have been times in the past where
revbump(1) has been telling lies and causing genuinely unrelated
packages to be bumped. It's good in general to crosscheck when using
it.

-- 
David A. Holland
dholland%netbsd.org@localhost