2009-08-02: Back from vacation
Evening clouds seen through the entrance archway of a Holiday Inn in Ann
Arbor, Michigan. This is where I was week before last for Cartel, a
conference of system and application administrators for a group of
universities who use similar underlying technology to solve similar
problems.
The last couple of weeks have been very busy. Two days after I got back
from Ann Arbor, two friends came from out of town for a visit that they
can only do once a year. They were here all last week, so every moment
that I could spare from work was spent with them, playing video and board
games and just talking. We made it until 4am one morning, and averaged
around 2am. It was exhausting, and I'm now running a sleep debt, but it
was absolutely wonderful.
But, now what I want more than anything else is to get back on a normal
schedule. Between one thing and another, I've not been on one for nearly
a month now. Today was spent on some cleanup around the house and a few
things I needed to take care of for work, and tonight I'm going to work on
the sleep debt. Tomorrow, I'll start resynchronizing my schedule and
re-establishing a balance between all the things I want to do. And, of
course, try to absorb all the great new ideas that I've gotten from long
conversations with friends. Oh, and get back into working from a to-do
list.
2009-08-03: Project anti-procrastination
I was going to post a different picture today for variety, but I put it
off until too late and have run out of time, so you instead get another
shot of the same evening sky as I posted yesterday, if from a different
angle and with different composition.
Today didn't work out entirely as planned, although not unsuccessfully.
Rather than getting back to zero inbox and properly on track with time
management, or working on finishing the packaging from last Friday, I
instead tackled the next hot work project. I just found out today that
AFS backup restructuring is the next thing on the hot list as I rotate off
the hot projects I'm finishing (the WebLogin confirmation screen removal
goes out tomorrow evening). So I started thinking about that while
walking to the post office to get my mail, realized that I already knew
how to implement it, and came back and wrote that down. Four hours of
work later, I'm finished with the first draft of the requirements and a
technical implementation plan that's sufficiently detailed that any Perl
programmer on the team should be able to finish the work.
There's still a lot of project left: allocating the storage, figuring out
if we're doing off-site storage replication, setting up a VM with proper
access to the storage, writing the code, writing a test plan and testing
various scenarios, and so forth. But most of that work can be done by
someone else. I suspect those four hours will end up being about half of
my total contribution to the project, and on the day that I was told to
start working on it. That's a good feeling, and will mean more time to
work on interesting things later.
(For those who are curious, we're getting away from backing up AFS via
Tivoli because the tapes and the Tivoli storage are extremely expensive
for reasons largely internal to Stanford. We get very, very few restore
requests for AFS, so due to that and due to other economics at Stanford,
doing vos dump full and incremental to a bunch of NFS-mounted disk that
can be replicated off-site ends up being very attractive for us. We're
probably going to replace our backup system with that.)
Tomorrow, I suspect I'll be working on the things I was going to work on
today: getting back into time management and finishing the packaging
work. Tomorrow evening, I do WebLogin upgrades, the confirmation page
removal, and KDC upgrades. Then the rest of the week looks nice and
open.
2009-08-04: RIP WebLogin confirm page
Skunk cabbage taken a year and a half ago in Nanaimo.
Today, I finished upgrading Stanford's WebLogin servers to WebAuth 3.6.1
and configuring them to suppress the confirmation page unless delegated
credentials were requested. This means that if one is already
authenticated to WebLogin, either by having logged on that day or by using
Negotiate-Auth to authenticate with Kerberos, one will be immediately
authenticated to Stanford web sites requiring authentication without any
intermediate notification.
Skunk cabbage is a good analogy for how I feel about removing the WebLogin
confirmation page. It does make things a bit quicker, but I think losing
the notification that one is going from unauthenticated to authenticated
space stinks. But it was an argument that I lost, and then we had a bit
of a fire drill to finish the confirmation page removal. So today I'm
feeling very relieved that it's off my plate and I can move on to work on
other things.
I still haven't caught up on the Debian Policy list, but I've now gotten
everything else out of my inbox. Digant and I finished "good enough for
now" Debian packages of Apache ActiveMQ today, so I can stop working on
that. Things are currently looking good for the rest of this week; I
should have a chance to work on some lower-priority projects that will be
fun.
2009-08-05: Forward momentum
It's been beautiful here, so I felt like posting something flower-like.
This actually isn't a flower; the flowers will eventually be in the
center, where there are only buds. This is the structure around a flower
cluster in a dogwood. The outside "petals" are actually leaves (bracts).
I had trouble getting moving today, which seems to be the standard problem
with Wednesdays. That apparently still needs some work. But I did push
through it and started on Shibboleth 2.2 packaging for Debian. The new
version of xml-security-c has been uploaded (and has already gone through
NEW!), and I have the work done for xmltooling and opensaml. I'll build
those tomorrow once xml-security-c has propagated into the archive and
upload them, and then start on opensaml2 and shibboleth-sp.
I also finally pushed myself to do the first step of the work to support
jaunty in our FAI build system. The FAI configuration is now managed via
Puppet rather than synchronized out of AFS, which will also make it easier
to maintain our disaster recovery site.
Given how many work hours I already have this week, I'm considering taking
parts of tomorrow and Friday to do Debian work, particularly Policy and
Lintian work since I'm rather behind on both.
Or maybe I'll do the Debian packaging of OpenAFS 1.5.61, since that was
just released.
2009-08-06: Push push push
Chain between a water control valve at Stanford. Despite the fact that I
put off doing this until way too late and am remarkably tired, I even
bothered to go into Gimp and rotate and crop it. I think this proves I'm
getting more comfortable with using Gimp to do some simple photo cleanup,
although I'm still way behind all the tricks that I know can be used.
Today required rather more effort to get started and keep moving, but I
still ended up being productive. The only major drawback to the day is
that I put off exercise until way too late and unless I develop a rather
strong second wind while writing this, I don't think I'm going to manage
to exercise today. Maybe I can manage to do so twice tomorrow... although
that usually doesn't work. Oh well, I get one day a week of slack, even
if I ideally don't use it.
I didn't manage to push Shibboleth packaging as far along today as I was
hoping, although that's partly because we decided to remove Shibboleth 1.x
from Debian and I filed those removal requests instead. I messed up the
first xml-security-c upload, so I've now uploaded that again and will
upload xmltooling tomorrow, probably. I also didn't get any Debian work
done, but that's because I started on kstart instead and kept working on
that until 22:30.
My Kerberos v4 Autoconf macros are now cleaned up and debugged (which took
nearly an hour all by itself, due to two missing double quotes), for
whatever good that will do going forward. But it makes me feel good.
I've also imported my minimal kafs library into rra-c-util and redid the
build system. I just need to write a test suite for it, and that's all of
my to-do items for the next kstart release.
I'm now over 40 hours of work for the week and I have to do e-mail
upgrades late tomorrow night, so the tentative plan is to spend all of
tomorrow except for the e-mail upgrades doing non-Stanford stuff, probably
Debian work. We'll see if that plan works out better than it did today!
2009-08-09: reminder 1.10
I noticed the other day while reviewing all of my reminders that listing
them in order by reminder number is not horribly useful. Particularly now
with reuse of reminder numbers, that order is essentially random. This
release instead orders reminders (for both active and list)
by the date at which the reminder came due.
You can get the current version from the reminder distribution page. New packages have also been uploaded to my
personal Debian repository.
2009-08-15: C TAP Harness 1.1
A while back, I felt like writing some code outside of any project I
needed to be working on and picked something off the C TAP Harness to-do
list. When running long test cases manually or with runtests -o,
it's nice to see a summary at the end to know whether any test cases
failed. This release adds exit handlers to the C and shell TAP libraries
to print out such a summary on exit, provided that all test results are
reported via the library calls.
This release also contains a couple of bug fixes, one for compatibility
with Perl's Test::More when skipping all tests and one to drain all output
from test cases that aborted to avoid spurious SIGPIPE.
With this release, Automake silent rule handling (new in 1.11) is now
optionally enabled and turned on by default for make warnings. I
love this. It makes it much easier to see the compiler output that I care
about.
You can get the latest version from the
C TAP Harness distribution page.
2009-08-15: rra-c-util 2.0
The largest change in this release of my C utility code, and the
justification for the version bump, is the addition of the portable
replacement for an AFS system call library, which I use for both
kstart and
pam-afs-session. I redid its
build system somewhat to make it more self-contained and added a test
suite relative to the version previously included in kstart.
There are lots of other changes in this version, particularly to the
Autoconf macros for Kerberos library probing and some of the macros that
underly it. There is a new mkstemp replacement for systems that don't
have it. I also reverted the separation of the message functions into
different object files depending on whether they can call exit.
Unfortunately, the latter change didn't do what I hoped; even if a library
doesn't use those functions, they're still linked in. So there's no point
in keeping the code duplication.
You can get the latest version from the rra-c-util distribution page.
2009-08-15: kstart 3.15
It's been more than a year since the previous release, so this was
long-overdue. Most of the hold-up has been converting the package to use
C TAP Harness and
rra-c-util, which required a lot of
resynchronization and led to the new releases of those packages.
The main new features in this release are that k5start and krenew catch
SIGALRM and immediately refresh the ticket cache even if it hasn't
apparently expired and krenew has a new option, -i, to continue
running even in the event of errors. Further improvements to error
handling in krenew and k5start are likely in the next release.
There are also a wide variety of more minor bug fixes and documentation
improvements. The build of k4start is now automatically skipped if no
Kerberos v4 libraries are detected. And, of course, there are all the
changes from updating to the current C TAP Harness and rra-c-util shared
code.
You can get the latest release from the kstart distribution page.
2009-08-15: Debian Policy 3.8.3.0
We'd accumulated quite a few fixes in the Debian Policy Git repository and
it seemed like it was about time for another release. I didn't get to
everything that I wanted to, and I'm not sure we're really keeping ahead
of the pace of new bugs, but it's nice to see a slightly faster release
cycle.
The summary of changes from 3.8.2 is:
-
Add DEB_*_ARCH_CPU and DEB_*_ARCH_OS variables and recommend them
over GNU-style variables for that information. [4.9]
-
Source package Architecture fields may contain "all" in combination
with other architectures. Clarify when "all" and "any" may be used
in different versions of the field. [5.6.8]
-
The Debian archive software does not support uploading to multiple
distributions with one *.changes file. [5.6.14]
-
The Binary field may span multiple lines. [5.6.19]
-
Remove the permission for shared library packages to install
libraries in a non-standard location and modify ld.so.conf.
Packages should either be installed in a standard library directory
or packages using them should be built with RPATH. [10.2]
-
Clarify installation directories for X programs and remove the
requirement to pre-depend on x11-common before installing into
/usr/include/X11 and /usr/lib/X11. [11.8.7]
-
Remove the requirement that all characters in a manual page be
representable in the legacy encoding for that language. [12.1]
-
Localized man pages should either be kept up-to-date with the
original version or warn that they're not up-to-date, either with
warning text or by showing missing or changed portions in the
original language. [12.1]
-
install-info is now handled via triggers so packages no longer need
to invoke it in maintainer scripts. Info documents should now have
directory sections and entries in the document. Packages
containing info documents should add a dependency to support
partial upgrades. [12.2]
-
The requirement for Perl modules to have a versioned Depend and
Build-Depend on perl >= 5.6.0-16 has been removed. [perl]
2009-08-15: Lintian 2.2.14
This release is primarily to synchronize with the Policy 3.8.3 release
that just went out, but we'd also accumulated some bug fixes and I did
about five hours of work this evening cleaning up the BTS and trying to
get the most obvious and easiest fixes in.
Lintian now has a new mechanism for diagnosing package dependencies on
metapackages (either regular or build) and I want to add to that a bit
more in the next release. Unfortunately, this is probably something where
we're going to have to hand-maintain a list of metapackages. If you have
a long-lived metapackage (a package that exists for the convenience of
users or installers to install a bunch of related things, but which
nothing should depend on directly) where you've had problems with invalid
dependencies on it, let us know so that we can add it to the list of
things Lintian looks for.
We're definitely not keeping up with the flow of incoming bugs at the
moment, although if I still had the time to devote a day every couple of
weeks on Lintian, we probably would be. I still need to do a BTS pass and
close some bugs that are invalid and reorganize things a bit, but tomorrow
is devoted to time away from programming. Five software releases in a day
is a lot.
I haven't updated the version of Lintian running on lintian.d.o for a
while. Monday is probably a good day to do that, unless something major
shows up in this release that needs a quick fix. Of course, I have a ton
of meetings on Monday, so we'll see if that happens.
2009-08-30: Term::ANSIColor 2.02
The substantive changes in this release are a new colorvalid function (not
exported by default) that checks whether attributes are recognized by the
module, and the addition of faint as a synonym for dark. The colorvalid
function replaces an earlier technique with catching exceptions at the
request of the Perl::Critic maintainers. (I think it's kind of silly, but
it was simple enough to provide.)
I took a look at running Perl::Critic over this module to see if it would
be useful, but my first impression is that it isn't particularly. Much of
the coding style suggestions are reasonable, but in some cases they don't
match mine or aren't useful for some reason, and all the references are to
a book that doesn't appear to be freely available. I was pondering
whether to add Perl::Critic to my normal set of test cases, but I think I
won't bother for right now. I have enough other things to work on.
You can get the latest version from the Term::ANSIColor distribution page.
All posts