On Mon, Mar 25, 2019, at 6:32 PM, Rich wrote:
> I'm not usually in favor of limiting configurability, but in this
> case, I agree with Joshua - letting people add their own custom
> defines for e.g. debian-buster or epel6 or ubuntu1604 might be fine
> (though at that point I'd still like something like a constraint
> solver to say "I want this to work on platforms X, Y, and Z" and have
> it DTRT, but if nobody else agrees that's probably overkill :P), but
> probably not a good plan to allow overriding portable{,-[0-9]+}, lest
> people be allowed to override what the consistent names mean and
> defeat the purpose.
>
> (I'm also torn because I somewhat feel like allowing overriding a
> subset of symbols might result in the same kind of problem, where e.g.
> every distro using ZoL 0.8 de facto ships a set of additional defines,
> but installing from source or running on e.g. illumos could result in
> inconsistent outcomes. But it's unclear how academic this fear is, and
> having some of the functionality is probably strictly better...)
>
By allowing a custom define you've now also added responsibility for having
that define available on every system that imports the pool. Which even if you
skate around the whole "this random system doesn't have the define used to
create the pool" and embed the define in the pool itself you have to handle
conflicts (what if the system already has a pool imported that has defined that
value differently?)
You quickly devolve to the moral equivalent of: "let's have a text file that
contains every IP and hostname on the internet and just ship that around. If a
host doesn't have the most recent version of hosts.txt then it won't work
correctly sometimes but that will be rare." :)
"No no, you're exaggerating, it's not like that at all."
I'd like to call the ball on this and say "Not for version 1.0. We are just
trying to improve what we have right now incrementally."
I'm not saying the request is without merit or value. We should tackle it at
some point. However doing this right needs a little seat time, which I think
we can get with data pools.
--
Thanks,
Josh Paetzel
------------------------------------------
openzfs: openzfs-developer
Permalink:
https://openzfs.topicbox.com/groups/developer/Td66d70a086ae5a9e-M3996bbdbc9e7c8c142c596cb
Delivery options: https://openzfs.topicbox.com/groups/developer/subscription
