It's officially a Windows porting month. Now that I'm half way through it
and with the last week of the month going to be a vacation, this makes
sense.

Today, finished up dealing with the timezone/timestamp issues on Windows.
This got stranger and stranger the closer I looked at it. After a timestamp
change, a program that was already running will see one timestamp, while a
program that is started after the change will see another one! My approach
works pretty much no matter how Windows goes insane though, and always
recovers a true timestamp. Yay.

Also fixed a regression test failure on Windows, which turned out to be
rooted in a bug in the command queue runner, which neglected to pass
along environment overrides on Windows.

Then I spent 5 hours tracking down a tricky 
test suite failure on Windows, which turned out to also
affect FAT and be a recent reversion that has as it's
root cause a [fun bug in git itself](http://marc.info/?l=git&m=140262402204212&w=2).
Put in a not very good workaround. Thank goodness for test suites!

Also got the arm autobuilder unstuck. Release tomorrow.