I've just installed Ubuntu 9.10 on my desktop machine (laptop). I have /home in a separate partition, so my usual habit is to destroy the old OS and install afresh. Most applications manage to do the right thing when they see my $HOME, as long as I keep the same uid.

Of course, before you do this, it's great to preserve a list of the extras that you had installed … dpkg will list them, but it formats the output for a tty even when being piped, so you have to trick it with the COLUMNS envariable to get full package names …

    $ COLUMNS=200 dpkg -l > ~/9.04.dpkg.list

Anyway, once the new OS is installed & running, grab the equivalent list and compare the two … we're interested in seeing any package that was in the 9.04 list that isn't in the 9.10 list …

    $ COLUMNS=200 dpkg -l > ~/9.10.dpkg.list
    $ for i in $(tail -n +6 9.04.dpkg.list | awk '{print $2}')
    > do
    > grep " $i " 9.10.dpkg.list >/dev/null || echo $i
    > done > 9.10.notin

That was 1009 packages on my machine, so I has a quick look through by eye and identified the best ones … many were libraries, so they could all be ignored. I got it down to 75 entries pretty quick. I edited 9.10.notin and deleted every line with an un-necessary entry, and then …

    $ sudo aptitude install $(cat 9.10.notin)

This makes sure that I don't forget about supertuxkart!

]]>

Announcing the first meeting of the Dunedin Skeptics in the Pub!

Come along over to The Duke of Wellington at 7.30pm on Wednesday 28 Oct to meet us all … http://dunedin.skepticsinthepub.net.nz/calendar/11698693/

]]>

Richard Stallman has recently conducted another interview with Kim Hill on Radio New Zealand; and once again I've produced a transcript – see /articles/rmsrnz2/index.html.

]]>

I've been doing some data validation recently, and this afternoon wanted to get a decent date field set up. I like ISO 8601 dates, they sort well in most default locales, and for the application in question I've gone for YYYY-MM-DD as the template.

This is trivially easy to detect as distinct from plain text, with

    \d{4}-\d{2}-\d{2}

but of course this does not do any validation of the content i.e. 9999−99-99 is seen as correct.

It's easy to improve on this expression to match the range of numbers for months 01 to 12 ⇒

    (0[1-9]|1[012])

, but actually making it correct is much harder. We need to know which months have only 30 days as a maximum, and to be able to detect leap-years that permit February to get to 29 days.

I had a quick google around – there are quite a few regexp cargo cult code libraries around the place, many don't match quite the same date format (yy/dd/mm is horrible), and most of they are simply not explained.

Then I found a nice article from Michael Ash (http://bit.ly/mashregex) that talked about ways to extend the simple versions, and explained the best bit, leap-year detection! I didn't need everything from his examples (I didn't want to make a leading zero optional, nor allow non-hyphen separators), and I only needed to validate from year 2000 to 2099, so I ended up with a much less complex expression.

The leap-year detection was easy enough, Michael had spotted a nice pattern that enumerates all the possible leap-years in a century (specifically, one where the initial year is also leap, which 2000 was) :-

00	04	08	12	16
20	24	28	32	36
40	44	48	52	56
60	64	68	72	76
80	84	88	92	96

This pattern reduces down to two parts –

    ([02468][048]|[13579][26])

The building blocks look like this :-

• Which months have 31 days in them? January, March, May, July, August, October and December
• So only these months can have “day 31” in any year
• “all day 31”

    (0[13578]|1[02])-31

• Which months have at least 30 days in them? Everything except February
• So only 11 months can have “day 30” or “day 29” in any year (i.e. whether the year is leap or not)
• “all days 29 or 30”

    (0[1,3-9]|1[012])-(30|29)

• All months can have days 01 through to 28, in any year
• “all days ≤ 28”

    (0[1-9]|1[012])-(0[1-9]|1[0-9]|2[0-8])

• February can have “day 29” if we are in a leap year
• “all Feb 29s”

    ([02468][048]|[13579][26])-02-29

Three of these four building blocks are valid for any year, and can therefore be grouped together. The pseudo-code looks like :-

• starts with “20”
• followed by leap-years, and is “Feb 29”
• OR
• followed by any two numbers (doesn't need to exclude leap-years)
• is any day ≤ 28
• OR
• is any valid day 29 or 30
• OR
• is any valid day 31

And the final expression (tested with Visual REGEXP, which is already packaged for Ubuntu as visual-regexp) ..

20(([02468][048]|[13579][26])-02-29|\d\d-((0[1-9]|1[012])-(0[1-9]|1[0-9]|2[0-8])|(0[1,3-9]|1[012])-(30|29)|(0[13579]|1[02])-31))

Formatted for readability:

20
(
    ([02468][048]|[13579][26])-02-29
    |
    \d\d-
    (
        (0[1-9]|1[012])-(0[1-9]|1[0-9]|2[0-8])
        |
        (0[1,3-9]|1[012])-(30|29)
        |
        (0[13579]|1[02])-31
    )
)

]]>

I've long had /etc checked in to version control on my machines, especially where there are multiple administrators – it makes life easier, having visibility of configuration changes!

I've just set up a new server, and this time I'm trying Mercurial as the versioning system, for the simple reason that it keeps the repository data in a single directory (unlike subversion).

Starting with instructions at http://tinyurl.com/mikas-hg, I came up with this :-

• Install the base OS
• Do any update / upgrade to get yourself current
• Install Mercurial
• Decide on the list of files in /etc that should not be versioned
• I tend to exclude things like SSL and SSH keys, as I often publish the repository later
• Add and check in!

root@box:~# aptitude install mercurial
...
The following NEW packages will be installed:
  mercurial mercurial-common{a} python-beaker{a} python-sqlalchemy{a} rcs{a}
...

root@box:~# cd /etc
root@box:/etc# hg init
root@box:/etc# chmod 700 .hg

root@box:/etc# cat > .hgignore
(^)*.dpkg-new
(^)*.dpkg-old
(^)blkid.tab(|.old)
(^)mtab
(^)adjtime
(^)ssl/private/*
(^)ssh/ssh*key
(^)ld.so.cache
(^)passwd-
(^)group-
(^)shadow-
(^)gshadow-

root@box:/etc# hg add
adding .hgignore
adding ...

root@box:/etc# hg ci -m "Initial post-install checkin"
No username found, using 'root@box' instead

My next step is generally to install Trac (with the mercurial plugin), and then I have a great platform for multiple administrators to co-operate, track changes, keep notes and raise tickets on each other!

]]>

I'm not especially happy with Ubuntu today.

It's probably nothing serious, and I'm sure I could work around it if I needed to, but …

The Ubuntu 8.10 Server installer won't run from a USB disk, despite the best efforts of Unetbootin and other more manual hackery. So I have to burn a CD …

The installer won't recognise my atl1e NIC, although the base OS does. So I have to do all the network config afterwards by hand :-)

And worse than that, Xen dom0 support has been removed. So if I want Xen I'll have to step outside the repository support and do it myself …

I did briefly try Debian Lenny, but couldn't get the Xen kernel to load with GRUB – my kernel seems to be “too big” and the bootloader won't install without a “BIOS Boot Partition” … and I can't find any way of creating one of those! That seems to be one of the punishments for using LVM on top of RAID in my boot partition :-)

]]>

I have a new lump of hardware on my desk, intended to be a household server. It's basically an Asus P5Q Pro motherboard, with a couple of SATA disks. With a few gig of RAM and a dual-core processor, it is intended to be a virtual machine server, running a firewall, MythTV/media archive, and some public sites.

VMWare ESXi was going to be the base OS; it's “free as in beer”, it's going to be used on some of the company systems I'm installing this year, it provides a nice and secure base OS and virtual networking that would make separation of different guests easy.

However, I didn't check the hardware compatability of ESXi before getting all enthuiastic. VMWare do use Linux under the hood, but I guess in the interests of reducing support issues, have only kept drivers for the most enterprise-class devices that they expect to see in the datacentre.

This means that although the P5Q has a supportable SATA device, by default ESXi doesn't recognise it. You have to hack both the ESXi installer, and the installed image to get it to work.

Also, ESXi does not recognise the onboard Atheros network interface, which means that it cannot be available for any of the guest machines. I would have to buy a supported NIC for my machine; that's basically an Intel one. It would have to be PCI-Express, as those are the only ports free. That puts the price up to around 40% of the entire of the rest of the server.

Sorry, ESXi 3.5. If you get wider hardware support, I may install you. Otherwise, it's off to Ubuntu and KVM …

]]>

Under NZ law (specifically, Section 92 of the Copyright Amendment Act), if your ISP receives multiple accusations that you have infringed copyright online, they will have to terminate your internet connection.

Or will they?

Now, this law (not all portions of which are actually in force today, but they will be soon barring a sudden change of heart by the Government) is widely regarded by the Internet industry as complete rubbish – technically unenforceable, morally wrong and even violation of human rights. Start reading at Creative Freedom NZ. But that doesn't stop it from being a law, does it?

Read the new juicy bits of law at http://www.legislation.govt.nz/act/public/2008⁄0027/latest/DLM1122643.html

An Internet service provider must adopt and reasonably implement a policy that provides for termination, in appropriate circumstances, of the account with that Internet service provider of a repeat infringer.

But what is an Internet service provider defined as?

http://www.legislation.govt.nz/act/public/1994⁄0143/latest/DLM345639.html#DLM1703710

Internet service provider means a person who does either or both of the following things: (a) offers the transmission, routing, or providing of connections for digital online communications, between or among points specified by a user, of material of the user's choosing: (b) hosts material on websites or other electronic retrieval systems that can be accessed by a user

Now here's the crux of my argument …

The person who is responsible for the account with “an ISP” (e.g. the person who pays the bill to connect their household computer to Xtra) is, under this Act, an ISP themself.

This is because it is now normal for Internet access to be available for every member of a household; therefore the person who “pays the bill” is providing connections to a user.

Seen the recent TV adverts for ADSL routers with wifi access? The adverts explicitly say that this machine should be used to enable multiple people in the house to access the Internet simultaneously. This is not some unusual tech-geek activity, it is shown to be normal practice for the sort of person who buys products advertised on prime-time TV – in other words, “everyone”.

So, if an ISP like Xtra terminate “an account” under the auspices of this law, I would argue that they have acted incorrectly; they have terminated access for an entire ISP and not for “a user”.

Just because “I as an ISP” have my network behind NAT, and only one IP address is visible to the outside world, is no reason to claim that I am not an ISP. If an ISP claims that their Terms and Conditions state that the connection should be used by only one person, I would argue that is an invalid restriction, and that their own industry advertises to encourage people to share connections.

I am an ISP. If a copyright holder wishes to complain about an action taken my a user of my network, they have to talk to me, not to my upstream ISP.

With apologies to Jonathan Swift :-

    Great ISPs have little ISPs downstream from them to bite 'em,
    and little ISPs have lesser ISPs, and so _ad infinitum_

]]>

I set up a new Unix box the other day – Tomato OS on a Linksys WRT54G. I was surprised to see that the time was set before 1970! But Unix time starts in 1 Jan 1970, doesn't it?

# date
Wed Dec 31 16:08:50 PST 1969

Mmm … it took a moment or two to sink in (actually, the evidence is that it took 31 seconds …)

# TZ=UTC date
Thu Jan  1 00:09:21 UTC 1970

Yep, Unix time starts on 1 Jan 1970 all right … in the UTC timezone. This machine had defaulted to PST, which is 8 hours west of the prime meridian …

Time Lords may relax; heavy transuranic elements will not be required.

]]>

I've had OpenWRT on my Linksys WRT54Gs for a year or so now, but although it works fine as an AP, I've not had much luck with WDS (Wireless Distribution System).

People have been recommending the Tomato firmware as an alternative for a while now, so I decided to give it a go. Mmm, a problem …

root@gl:~# mtd -r write WRT54G_WRT54GL.bin linux
Bad trx header
This is not the correct file format; refusing to flash.
Please specify the correct file or use -f to force.
Image check failed.

Unfortunately, the Tomato firmware is distributed as ‘.bin' files, not ’.trx', and according to comments on the x-wrt website (http://forum.x-wrt.org/index.php?topic=429.0;wap2) the OpenWRT flash updater mtd no longer supports these for installation, so you can't get there from here …

The FreeWRT mailing list (http://osdir.com/ml/embedded.freewrt.general/2007−01/msg00007.html) claims that their mtd binary can be used for the install, so I tried that …

root@gl:~# ./mtd-static -r write WRT54G_WRT54GL.bin linux
Unlocking linux ...
Writing from WRT54G_WRT54GL.bin to linux ...  [w]

Rebooting ...

Unfortunately, the machine ended up bricked. Actually, both a WRT54GL and a WRT54Gv2 were bricked by this. Luckily, it was only a bad boot loader, and the dd-wrt wiki has a great page for diagnosis and recovery actions – http://dd-wrt.com/wiki/index.php/Recover_from_a_Bad_Flash

So in the end, I just used the firmware's default TFTP-on-boot facility to put the Tomato OS onto the devices, which worked just fine. And without getting involved in huge configuration, WDS is now operating between the ISP-connected unit, and the one with all the wired devices connected; and they both work as Access Points too.

]]>