On Tue, Sep 28, 2010 at 11:24 PM, Kyle Markley <[email protected]> wrote: > > Tonight I finally got around to upgrading my grid to 1.8.0 and it went > smoothly everywhere except on Win7-64. There is something wrong with the > dependency checking there; setup.py build ran cleanly but setup.py test > complained that the installed version of pycryptopp was too old. (It's > right, it is.) I reran setup.py build and saw no mention at all of > pycryptopp -- it appears that somehow it wasn't even checked, so the old > version wasn't detected until setup.py test.
"python setup.py build" doesn't check for run-time dependencies (packages that are needed to run the code), only for build-time dependencies (packages that are needed to build the code). > Two other systems (WinXP-32 and Vista-32) didn't have any problem with the > upgrade, so this may be a platform-specific problem. Perhaps they didn't have a too-old version of pycryptopp installed? > Perhaps relatedly, a few months ago I noticed an attempted -SUMO build > download dependencies from the internet instead of using the local copies. If you could track down more information about this we could try to fix it. If it was a snapshot instead of a stable release then it could have come from the period where we had increased the dependencies but hadn't yet included the new packages in tahoe-deps/ and SUMO. If it was a stable release then I am surprised that it did that. > Is dependency checking in setup.py build known to be a bug-rich area? Are > there any tests that cover this? Yes and yes. The tests include the "Desert Island test" on the buildbot, which goes red whenever a SUMO build attempts to download something: http://tahoe-lafs.org/buildbot/builders/clean That test *was* red for a while due to the aforementioned period of having insufficient SUMO. Regards, Zooko _______________________________________________ tahoe-dev mailing list [email protected] http://tahoe-lafs.org/cgi-bin/mailman/listinfo/tahoe-dev
