Now, django.contrib.admin was in settings.INSTALLED_APPS, and django.template.loaders.app_directories.Loader was in settings.TEMPLATE_LOADERS.

So, why was django throwing an exception? To find out, I stepped into a shell:

>>> from django.template.loader import find_template_loader
>>> loader = find_template_loader('django.template.loaders.app_directories.Loader')
>>> loader.load_template('admin/login.html')

This was where I realised something was wrong. I don’t get the error now (as I have fixed it), but it complained about being not allowed to open the file. As in a permissions error.

Looking up the location in a new shell, I was able to see that all of the files in the django.contrib.admin.templates directory were only able to be read by root. For some reason, python setup.py install had set the mode of these files to 0600, instead of the expected 0644. A quick sudo chmod -r ag+r templates (from inside the django.contrib.admin directory) fixed it.

I also use the Programmers Dvorak layout, including swapping the orientation of the numeric keypad to be more like a phone, not a calculator. This was the only part I needed to actually swap the keycaps on my keyboards around on, as I touch-type on the main keyboard, but tend to look a bit more when entering values.

I have noticed that quite a few applications don’t handle a different keymap that well. It appears they are using the keycodes, which are independent of the values that should be presented to the operating system. I can understand why this is in many cases: for games for instance, you want to be able to continue to walk around using the aswd keys in World of Warcraft regardless of the layout of the keymap.

I guess it makes a bit of sense to pass through this stuff to the virtual machine too: after all, it has its own keymap to interpret. With VMWare Fusion, however, the translation of Mac Shortcuts to Windows Shortcuts is slightly broken. Because on my keymap, the ⌘-keys match their keycode (didn’t want to re-learn all of the shortcut keys), then they are not passed through correctly.

Luckily, they are editable in the VMWare Fusion preferences.

Mine now look like this:

Now, I can hit ⌘S in notepad, under Windows, and it saves my document. And I can use the cut/copy/paste/undo shortcuts again!

However, I have many aliases and functions that need to act differently if I only have a console session at the machine in question, or a full GUI session.

With bash, this is easy to test:

export EDITOR='nano'
if [[ -z "$SSH_CONNECTION" && $OSTYPE =~ ^darwin ]]; then
export EDITOR='mate --wait'
export TEXEDIT='mate -w -l %d "%s"'
export LESSEDIT='mate -l %lm %f'
fi

Now, if I am remotely connected to a machine, then I will get nano as my editor, but if I am sitting directly in front of it, then it will open Textmate.

I made a django app that provides a new management command: bshell. This will start a new shell, using bpython, and import all of your models.

You can get it with:

pip install django-bshell

And then install it into your django settings.INSTALLED apps. The app itself is called ‘bshell’. Then you can just use:

django-admin.py bshell

The code can be found on bitbucket.

I disabled IPv6, and they all cleared up.

I don’t know if this was an issue with Internode’s IPv6 stuff, but it was being handled by my Airport Extreme. My iPhone was working fine, because it doesn’t use IPv6.

What a relief.

I had tried everything I could think of: from changing my DNS server, DHCP server, I even tried a reinstall of my OS (although this wasn’t why I did that - I wanted a cleanup of my dev machine).

It now handles running tests when (a) The apps are not directly in the project root, but inside another folder, for instance; and (b) the app/tests.py file has been split into seperate files.

The main reason I made this was so that I could run tests and have clickable links in the results window for the location of failing tests.

There is still much to do on this. I am considering re-writing it in python rather than ruby, so I can programmatically find the app name, rather than guess it. I also want to refactor the hell out of it and make it much nicer.

Anyway, if you are interested, you can find the most recent version at http://github.com/schinckel/Django-Management.tmbundle - and I think it also appears in TextMate’s getBundles bundle.

Clicking in the search box results in:

So, there is an image in the placeholder. This is very neat!

Unfortunately, it appears this might not be a part of the HTML5 spec, but is done using JavaScript.

Today, I noticed a Copy button at the bottom of the window:

This still requires you to enter your keychain password to authorise the copy, but saves a step or two.

This goes even further, with occurrences that touch or overlap being merged into potentially a new event. This is a requirement of our problem domain, since these events are used to see if a person is available to perform a task, and two adjacent occurrences would otherwise make it hard to determine if the person is indeed available.

So, it turns out that this is only part of the problem. In addition, changes to availability may require approval by a manager. So, in addition to being able to merge objects, we also need to be able to merge pending objects with one another, and mark pending objects as superseding other objects. And, even more fun, if a pending object is created that matches a pending-deletion object, for instance, then that object is restored: the two pending events cancel one another out. Finally, a pending object can also be ‘psuedo-merged’ with an availability, where it then is the union of the two events, and supersedes the approved event!

This problem is even stickier than the previous part. Having said that, I have it working, although at this stage it is not possible to have recurring pending events. Things were made even more difficult by the way that I handle superseding events. This cannot be just references in the database, as events and occurrences may be deleted and created by saving another event, so a different method of keeping track of which events are superseded is required.

The good news is, this is coming along well. It did occur to me on the way home today that without ‘never ending’ events, this would possibly collapse the problem space down. Occurrences would no longer need to be recreated quite as often, and indeed they may be the sole storage representation: abstract events could then be created on the fly, purely for descriptive purposes. All I need is some way to represent events that don’t have an end date…

…or I could make it that never-ending events must end in 2050. Then, any event that ends that year will be represented to the user as never-ending. This seems like a slippery Y2K-style slope, but if I keep moving the goal-posts, then it might work. And simplify my code immensely. It turns out there are only ~2100 weeks until then, and my events at this stage only occur weekly (it makes the simplification simpler to only have to look for weekly patterns, and that is how people usually determine their schedule).

I’m glad we had this little chat.

There are really only two ways that events can be merged. If two events have the same frequency, and occur on the same day, or on consecutive days according to the period, and they have the same start and finish times, then they can be merged into one.

Similarly, if two events have the same period, occur on all of the same days, and overlap or touch in times, they can be merged into a single event.

From there, it starts to get complicated. Two events that partially overlap may be able to merge parts of themselves, possibly resulting in an extra event being created.

For instance, an event every Tuesday at 9am-12noon, that lasts for 10 weeks, and another event every Tuesday 11am-2pm that starts 5 weeks later, and also runs for 10 weeks, will result in 3 events: one that lasts 5 weeks 9-12, one that lasts 5 more weeks 9-2, and one that lasts a further 5 weeks 11-2.

This on its own is easy to deal with, but as soon as you consider events that have no end date, things get tricky quickly.