Re: better changeset tracking

Hi Wez,

As you said, it's pretty straightforward to handle bugs this way, but a pain to open a ticket for every little maintenance job. We solve that problem by opening up a maintenance ticket per milestone (eg: X.Y.Z release maintenance ticket) and use that as a catch all for those misc commits. When we're done with working on a ticket in trunk, we set its priority to merge so that we know that it is ready to be considered for merging down to the stable branch(es), and we close it only when that merging is complete.

This could work, except of course we don't have any such thing as 'a maintenance ticket' or a way to set priority to 'merge'. It's kind of the opposite way around to the way the PHP bug system works... and it probably would be a pain to have it as part of an open system (in the sense that users can add to it). Perhaps adapt the bugs *interface* to deal with maintenance tickets along the lines you suggest, but with a separate db and ID system - so we get #M9999 rather than #9999, and 'merge', 'no-merge', 'close' as the options?

This process should be pretty easy to translate to php.net development. I'm not saying that this is the one true way to do development, but it works for us at OmniTI.

It's the difference between a controlled environment and open source.... but I can see it working if the 'per milestone' route is taken, that's a good idea. Anyone with QA/bugs karma care to intervene at this point? - Steph

--Wez. On 5/28/07, Steph Fox <steph@zend.com> wrote:

Hi Wez,

I think the key is in forcing every commit to reference a ticket; that way you can't "lose" a changeset by forgetting to put a bug number in there.

Going back on-list... It's a good idea in principle, but I can foresee some problems with it. Many (most?) of the unnumbered fixes Ilia's been making have been alterations to internal string or memory function calls. Generally those are either performance or security changes, replicated throughout the code base and starting with a single 'blast' aimed at upgrading everything, followed by lots of smaller catch-up fixes over the next several weeks. This kind of thing isn't really worthy of a bug report, unless you have an entirely new concept of what a bug might be and allow for change types like 'maintenance', 'security' or 'performance' via the bug database somehow. The nearest we've come to that recently is MOPB - everyone marked those fixes as such - and occasionally coverity. If the bug db allows it it might make sense to put an initial function change as a 'maintenance' report with some explanation and a number, e.g.:

helly           Sat Feb 24 18:20:46 2007 UTC

 Modified files:              (Branch: PHP_5_2)
   /php-src/main       snprintf.c snprintf.h

Log: - Add [v]slprintf to overcome the returnvalues issue of snprintf would get a fuller explanation on the report, and every subsequent "snprintf() -> slprintf()" commit would reference that maintenance report number. Fine, but what to do about "malloc() -> pemalloc()"? Would all 'maintenance' fixes require a test script? Wouldn't the need to check for a maintenance number put devs off making a simple one-line fix? Branch-specific fixes are generally (but not always) marked as such at present. People tend to mention it in the 5_2 branch but not in 4_4. - Steph