Let me give you the long version, it'll make me feel better :)
I upgraded to OS X Snow Leopard on Friday. It mostly went smoothly,
but I did have some Eclipse problems at first. This turned out to be
related to the 64-bit transition and when I installed the 64-bit
version of Eclipse all seemed OK.
For my main Subversion 1.6.x client, I use the binaries that Jeremy
provides on openCollabNet (which includes 64-bit support). I also
built myself as part of debugging this.
Today I noticed in Eclipse History view that I could not see the
affected paths for any revisions. When I debugged it came back null
from the JavaHL API call. I IM'd some other OSX-using developers to
make sure it was not a bug in latest binaries or latest Eclipse/
Subclipse. Just me. So I assumed the bug was Snow Leopard or
something with 64-bit. Somewhere in the process of trying to write
some new JavaHL tests to show the problem my Eclipse started crashing
right and left inside JavaHL library.
It occurred to me, once I got Eclipse somewhat stable again, to try a
file:// repository (because tests were passing). This worked. At
this point, a lightbulb went off. Also on last Friday I changed my
~/.subversion/servers file to use Serf so that I could test a problem
that Edmund was having on trunk. This turned out to be the problem.
Putting this back to Neon solved all of the problems.
At a minimum, I think that the JavaHL calls to our log fetching API's
is not returning the paths via Serf. Those of us that hack on JavaHL
(as well as the buildbots) only run the tests over ra_local. I think
that dlr (or maybe someone else) made it possible to run the JavaHL
tests via other layers. I have no idea how to do this though.
Received on 2009-09-01 23:37:20 CEST