[Ns-developers] Request to commit -fvisbility=hidden pythonbindings patch

[Gustavo Carneiro]

2009/3/3 <craigdo at ee.washington.edu>

>
> > BTW, I guess the result of yesterday's discussion regarding regression
> > testing / bug 370 is that maintainers DON'T need permission to push
> > changes during this phase.  We will however, transition to that mode
> > of operation at some point in the future, say the last week or two of
> > this month
>
> I found there were two periods when it was extremely helpful to lock down
> the dev repo and run *all* the tests before continuing.  It's not as
> critical at other times ...


Craig, I don't understand what you mean by "*all* the tests".  We have
automated regression testing on several platforms and compilers.  What more
can we do manually, and why can't that manual work be done automatically by
the build farm?


>
>
> 1) When major merges were happening -- I called this merge week.  We really
> want to demonstrate that a new feature is going run on all of our supported
> platforms (yes, including Cygwin and osx ppc, not just on a subset of
> compilers).   We want to make sure that new bits will run properly and not
> break existing code.  It's really impossible to determine this against a
> moving target.  I had times in which virtually everyone was committing
> changes to this or that module while we were trying to merge something new
> in.  It's extremely painful at a minimum.  I believe you have had the
> pleasure of experiencing this kind of thing, Raj :-)
>
> 2) During the code freeze/RC period.  We are really striving for stability
> here.  It is a really, really bad sign for the system to be broken in any
> way during this time.  It would be bad for RC-n to come due and to have the
> release manager run the tests and discover that the system was broken the
> week before by a "bug fix."  I think the release manager really needs help
> from the developers to make sure this doesn't happen; and I think we should
> start helping out perhaps a week before RC1 by starting to run all of these
> tests when we make changes.  More testing eyes make for less catastrophic
> problems, so to speak.


Since we have regression testing, perhaps what we need is just to add more
regression tests, as needed, not manually test things before releases.  The
main advantage of nightly regression tests is that you find out the next day
if your commit broke something.  If you only test for releases then it
becomes much harder to determine what commit broke things, and how to fix.

-- 
Gustavo J. A. M. Carneiro
INESC Porto, Telecommunications and Multimedia Unit
"The universe is always one step beyond logic." -- Frank Herbert