We've seen a few cases where the test suite for the ruby bindings fails
spuriously. There are two (separate) issues:
a) there is a race between svnserve failing in bind() because the chosen
port is already in use, and the test suite checking that the forked
svnserve process is still running.
b) the set of ports used is not randomized so concurrent runs on the
same machine will be fighting over the same ports
Is something like the patch below ok? It does not guarantee success but
makes failure less likely.
(p.s. congrats to all for the 1.8.0 release ;)
[[[
* subversion/bindings/swig/ruby/test/util.rb
(SvnTestUtil#setup_default_variables): Randomize port range used
for svnserve.
(Svnserve#setup_svnserve): Increase chance of catching an svnserve
failure.
]]]
Index: subversion/bindings/swig/ruby/test/util.rb
===================================================================
--- subversion/bindings/swig/ruby/test/util.rb (revision 1494521)
+++ subversion/bindings/swig/ruby/test/util.rb (working copy)
@@ -73,7 +73,8 @@
@realm = "sample realm"
@svnserve_host = "127.0.0.1"
- @svnserve_ports = (64152..64282).collect{|x| x.to_s}
+ sport = (50000 + rand(100) * 100)
+ @svnserve_ports = (sport..sport + 99).collect{|x| x.to_s}
@tmp_path = Dir.mktmpdir
@wc_path = File.join(@tmp_path, "wc")
@@ -286,6 +287,8 @@
"--listen-port", port,
"-d", "--foreground")
}
+ # wait a while for svnserve to attempt a bind() and possibly fail
+ sleep(1)
pid, status = Process.waitpid2(@svnserve_pid, Process::WNOHANG)
if status and status.exited?
if $DEBUG
