[Mondrian] RE: Test failures with high-cardinality dimensions

John V. Sichi jsichi at gmail.com
Tue Jul 15 01:35:10 EDT 2008

Julian Hyde wrote:
> Looks like we have very different opinions on this issue. The way I see 
> it, we have a half implemented feature. I think that most mondrian users 
> would see it that way too.
> (I'd like to hear about other developers on the list feel about this.)

One of the nice things about open source is that it doesn't always have 
to be either/or; it's possible to live happily in quantum superposition 
for quite a long time, with a suitable approach to source control.  Take 
a look at the way Linux kernel development works.  There's a primary 
trunk which everyone wants to get their stuff into eventually, but there 
are a lot of review/quality barriers to prevent half-working stuff from 
getting in.  Then (besides stable/release trees) there are many 
experimental trees.  Projects like User Mode Linux and uCLinux spent 
years as experimental trees, with lots of changes all across the kernel. 
  The maintainers had to keep merging down everyone else's work from the 
trunk, and had to keep submitting their own updated patches back 
upstream; through enough perseverance, they finally got their work 
included into mainstream kernels.

Experiments which lack dedicated maintainers will die a natural death 
without ever doing any harm to the trunk, since they never got into it 
in the first place.  And even then, they may remain useful as salvage 
yard for later attempts at similar features.  The merge burden is placed 
on the experimenters (rather than on people maintaining features already 
in the trunk), which is as it should be.

This approach requires heavy use of private branching, which is 
something that Julian has discouraged so far, for good reason (it's nice 
to be able to see the history of a file all in one place, without 
chasing changes across integrations).  It also requires stringent code 
review practices, which not everyone has time for.  A comprehensive 
performance regression suite like we've been discussing is also a 
must-have as a barrier.  But quality never comes cheap.


More information about the Mondrian mailing list