> However, for development teams with enough resources to outrun upstream projects - to develop their own stuff and to see specific benefits faster than waiting for such things to appear upstream, if ever - it becomes tempting to fork the code and ignore the community.
And it is also tempting since it also isolates you from the opposite problem: inconsiderate upgrades causing regressions and lack of backward compatibility. Honestly, which distribution cares about backward compatibility from one release to another? Just for fun compare with Windows. Or even worse, think about Fedora who keeps issuing hundreds of non-security upgrades after the release. And this not just on the latest release but even for older "releases".
I realize backward compatibility is expensive while most Linux distributions are free. But still: if you do not care about external developers then you cannot ask them to behave and not fork.
(RedHat is off topic here since we are talking about a web browser)
Posted Dec 9, 2009 9:56 UTC (Wed) by mpr22 (subscriber, #60784)
[Link]
It's amusing that you mention Windows on the backwards-compatibility score, when just about every Windows application I've ever installed appears to follow the "big ball of binary blobs" approach.
Needless Forks
Posted Dec 10, 2009 0:16 UTC (Thu) by marcH (subscriber, #57642)
[Link]
Even on an underlying OS that actually cares about stability you can still find some independent developers doing "the wrong thing". So guess how much chance there is of independent developers ever relying on system librairies provided by Linux distributions which do not care about stability at all? None.
Needless Forks
Posted Dec 10, 2009 0:49 UTC (Thu) by rahulsundaram (subscriber, #21946)
[Link]
Yet, that is what is happening a lot more than before. The problem is not that distributions do not care about it (although the level will vary) but a lot of upstream projects have historically ignored this problem. Projects like GTK have been exceptions. A lot of base platform libraries are adopting a saner model (as defined in http://www106.pair.com/rhp/parallel.html) over time and distributions will naturally inherit this trait. The alternative is just sticking to a base version and backporting which has its own tradeoffs.