|
|
Boost Testing : |
From: Aleksey Gurtovoy (agurtovoy_at_[hidden])
Date: 2005-07-12 07:57:54
Douglas Gregor writes:
> We're going to have to branch for release some time soon, to isolate
> the upcoming release from accidental changes and such.
I'd prefer to branch only when we are ready to prepare the release
tarballs: there is an overhead for both fixing things on the branch
and switching testing to the branch, and, furthermore, we need to
decide whether we want to test on the branch at all or we should rather
test the tarballs themselves (or both).
> Can we smoothly move all of the regression testers over to the
> branch?
Sure.
> What would be involved?
Martin described most of it already, but I guess it wouldn't hurt to
reiterate:
1) Regression runners need to add "--tag=RC_1_33_0" option to their
'regression.py/collect_and_upload_logs.py' command line
2) The hourly tarball maker needs to be switched to the branch.
3) The report generation needs to be switched to the branch. The
correpsonding reports will be available at
http://engineering.meta-comm.com/boost-regression/RC_1_33_0/
Note that all the above assumes the we will be testing _the release
branch_. Tarballs would require a slighly different setup.
>
> I'm imagining that I can change the default value of "--tar" from
> CVS-HEAD to RC_1_33_0 in regression.py, and after a full cycle of the
> regression tests everyone using regression.py (is that everyone?) will
> have switched and we'll be testing what we want to test.
I'd prefer to restrain from modifying 'regression.py', although I do
sympathize with the idea of automatic, centralized switching everyone
to another branch/source without having to ask them to modify their
scripts/command line. We were thinking of making 'regression.py' to
retrieve the current tag (if it's not explicitly provided in the
command line, that is) from, let's say, Boost SourceForge http area,
but other suggestions are most welcome.
> However, I
> expect that there's something that will need to be done on MetaComm's
> side so that we can see these results.
Just 2) and 3) above.
-- Aleksey Gurtovoy MetaCommunications Engineering