http://buildbot.darcs.net/buildbot-darcs/builders/zooko windows
Windows-XP i386/builds/57/steps/test/logs/stdio

Running shell tests on hashed repositories...
-rThe system cannot find the file specified.

All tests successful!
Done running tests on hashed repositories...
Running shell tests on old-fashioned repositories...
-rThe system cannot find the file specified.

I'm working on this bug.  I think that what is going on is that the rules to
implement the "make test" targets depend on a lot of unix-like functionality, e.g.:

test_shell_hashed:     darcs
	$(call shell_harness,hashed,\
	 echo ALL --hashed >> .darcs/defaults,cat)

Where shell_harness is:

shell_harness=$(call harness,$1,$2,$3,shell,sh,$(SHELL_HARNESS))

where harness is:

define harness
	@echo Running $4 tests on $1 repositories...
	@rm -rf tests-$4-$1.dir && cp -R tests tests-$4-$1.dir
	@cd tests-$4-$1.dir && rm -rf .darcs; mkdir .darcs;\
	 $2;\
	 if [ -r $(TEST_FILTER_FILE) ];\
	 then grep "\.$5" $(TEST_FILTER_FILE) | grep -v '^#' | $3 | xargs $6; \
	 else ls *.$5 | sort -r | $3 | xargs $6; fi
	@echo Done running tests on $1 repositories...
endef

This is okay in principle -- it just means we require cygwin to run the tests,
which is fine with me -- but I need to figure out what part of it isn't working
on Windows...

Okay, now it is at least running well enough to start the test script.  The
trick was to change my Windows "%path%" environment variable to put
c:\cygwin\usr\bin earlier than c:\WINDOWS\System32, so that the "sort.exe" that
gets called by GNU make when it wants to launch a test run is unix sort.exe and
not Windows sort.exe.

Now there are a bunch of test failures, some of which no doubt need to be fixed
"on my end", such as further configuration of my %path%, others of which no
doubt need to be fixed "on the darcs end", such as by changing darcs to work
better on Windows.

That's my assumption anyway.  We'll see...

Thanks to kowey and others, this has been fixed!