|Just another Perl shrine|
Re^4: Small incremental releases are good!by adrianh (Chancellor)
|on Feb 09, 2004 at 10:59 UTC||Need Help??|
Nooooooooooo! If you change your code, then you need to test all the parts it touches.
As I said before:
Just because there are regular updates doesn't mean you need to use them. I don't go around every live project and update modules every time a new release comes out. Release and integration are independent tasks.
Just because there are many releases doesn't mean you have to integrate every one. If you need the features / bug fixes you integrate. If you don't wait. They are independent tasks and can be scheduled appropriately.
That said, I would still recommend integrating as often as possible for the other reasons mentioned.
That means every nook and crany has to be looked at over and over. A big PITA.
As I said before:
If integrating code is causing problems spend some time figuring out why. You can often make it easier
This is what regression tests, acceptance tests, etc. are for. I run all my integration tests several times a day. Microsoft build their apps daily.
Obviously before a full public release of a public application you'll go through a slightly more extensive testing period. However if you're QA group cannot integrate minor updates in less than a week you should buy them some books and get some consultants in to train them up in some more modern practices.
If you can accumulate an amount of bugs over a month or over 30 seperate days, fixing them on schedule, you fix the same amount of bugs
But do you fix the same bugs? If you release more often you can re-order your priorities more often and respond to changes more quickly.
If a release takes QA a week to test minimally, trying to squash it to a day or two only hurts them.
In my experience when somebody says "We can only integrate once a month because it takes a week to do", it turns out in the end to mean "It takes a week to integrate because we only do it once a month".
An integration and testing process that happens once a month is built up around bulk changes, testing many new features, dealing with many integration failures, feeding large amounts of feedback back to the developer, etc.
If you do it more often you're dealing with fewer changes, fewer features, fewer integration failures, less information, tighter feedback loops, etc.
In most cases the more often you do it the faster it gets.