On Tue, Jul 26, 2011 at 2:08 PM, Daniel Friesen
<[email protected]> wrote:
> RequestContext WAS an RFC.
> No one commented on it...
>

An e-mail to the list announcing the RfC would've been nice ;-) Also
waiting for more than 1 other person to edit it would've been nice too.
If you're not getting responses: pester people for them.

> When it was originally committed it was a small change. It just grew as
> people improved it.
>

It started out small enough, yes...but it quickly grew to permeate a good
chunk of core code. And then the architecture was changed halfway through,
leading to the crap with __get(). Having more eyes on the RfC would've
avoided this, IMHO.

> Frankly if RequestContext and features of similar size had been
> committed to a branch instead of core, they would have bit-rotted there
> and never made it into core. Not to mention the relative size compared
> to things like ResourceLoader and Installer makes the negatives of svn
> branches relatively worse. Dozens of med-small changes getting shoved
> into branches and atempted to be merged into core?
>

Medium to small changes should still be done in trunk, and at no point in
my original e-mail did I say otherwise. What I'm asking for here is for large
refactors/rearchitecturings to be done in branches. RequestContext may
not have *started out* as a large refactor, but the architecture implications
are huge in the long run.

> I'd love to do work in branches, have everyone actually try the branches
> out and add their improvements there, then get them reviewed into trunk.
> But that workflow... that's really git, not this svn mess we have now.
>

Let's please not turn this into a Git v. SVN debate again. Git is cool, and
will perhaps solve some of our problems, but it's not a magic bullet to
review/merging. I believe first and foremost the debt of currently unreviewed
code should be near-zero before we even consider a migration. It would be
very easy to work ourselves into a very similar situation using Git (tons of
unreviewed pull requests--waiting in their branches for some love), and
anyone who thinks we couldn't get backlogged in Git is kidding themselves.

> Even if we kept this in the environment of svn and tried to use RFCs...
> people would have to actually discuss those RFCs... looking over the
> RFCs we have it looks like only 10% of them even get anyone discussing
> them. The rest just bit rot... So we're supposed to move from having a
> volunteer willing to code a feature and putting it into trunk, to having
> them putting up an RFC or putting it into a branch, and letting the code
> bit rot or RFC just sit there forever, and neither ever get into core?
>

I would rather people take the time to do it right than have a constantly
unstable trunk that makes backporting damn near impossible. For example,
you landing your Skin overhauls a day or two after branching REL1_17 was
a poor decision on your part...and made backporting stuff and releasing 1.17
a huge pain.

This kind of comes back to the idea of code freezes (boo, I know...). I don't
want to ever completely freeze trunk; that discourages contributions as we've
discussed before. I'm just asking for a little bit of common sense from all
around so we can keep trunk stable and push for more regular deploys and
releases.

-Chad

_______________________________________________
Wikitech-l mailing list
[email protected]
https://lists.wikimedia.org/mailman/listinfo/wikitech-l

Reply via email to