[OmniOS-discuss] header/library mismatch

Josef 'Jeff' Sipek jeffpc at josefsipek.net
Fri Mar 18 16:00:23 UTC 2016


On Fri, Mar 18, 2016 at 11:23:38 -0400, Dan McDonald wrote:
> 
> > On Mar 18, 2016, at 11:12 AM, Josef 'Jeff' Sipek <jeffpc at josefsipek.net> wrote:
> > 
> > I think it'd be fair to include a dependency in
> > pkg:/system/library restrict which combo of headers & libc can be installed.
> > E.g., something like:
> > 
> > 	depend type=incorporate fmri=pkg:/system/header at 0.5.11-0.151016:2016...
> > 
> > Dan & other IPS gurus: any thoughts about this?
> 
> I'm wary of incorporate dependencies because of how they've been abused in
> non-gate IPS packages.  Often those incorporate dependencies block
> upgrades.

Yeah, incorporate dependencies have been abused by userland/etc.  For this
libc+header case specifically, of all the types of dependencies incorporate
is the right one - this specific version, no more, no less.  (Other types
allow the depend target to be newer than the specified version which is what
this would try to avoid.)

> OTOH, system/library and system/header are peas in a pod, and live in the
> same build space, so it's possible this would be a good idea.

I actually think it would be a very good idea otherwise it's quite possible
to end up with a mismatch between libc & the headers.  If you are lucky, the
user that runs into it figures it out.  If you are unlucky, you'll get
sucked in and likely waste a bit of time.  Either way a simple dependency
would have made it impossible to end up with the wrong package combo.

Or did I misunderstand and mismatched libc & headers are supported by
OmniOS? ;)

> Someone should experiment with that.  I more than a coinflip's certain,
> however, it won't be making it into r151018. (i.e. Someone will have to
> strongly convince me otherwise.)

That's fine.  I could see this becoming a thrill-seeker^W^Wbite-sized bug.

Jeff.

-- 
Humans were created by water to transport it upward.


More information about the OmniOS-discuss mailing list