Goodbye Taskwarrior
It was good once, but it's time for a different task-tracking tool.
| Category | Data |
|---|---|
| Pending | 328 |
| Waiting | 10 |
| Recurring | 3 |
| Completed | 7852 |
| Deleted | 1768 |
| Total | 9961 |
| Annotations | 690 |
| Unique tags | 137 |
| Projects | 197 |
| Blocked tasks | 107 |
| Blocking tasks | 86 |
| Undo transactions | 3 |
| Sync backlog transactions | 12 |
| Tasks tagged | 46.4% |
| Oldest task | 2018-11-01 |
| Newest task | 2026-11-14 |
| Task used for | 8.0y |
| Task added every | 7h |
| Task completed every | 8h |
| Task deleted every | 1d |
| Average time pending | 8w |
| Average desc length | 21 characters |
Eight years ago, tasks were held as a bunch of json text, like this, but on
one line.
1{
2 "id": 0,
3 "description": "place new icons",
4 "end": "20241216T002703Z",
5 "entry": "20241215T215658Z",
6 "project": "splint",
7 "status": "completed",
8 "uuid": "1d9d7cb5-67ea-41fd-bafc-795c3e74806e",
9 "tags": [
10 "web"
11 ],
12 "depends": [
13 "e082f546-fef9-4a3d-b05e-e30d22e5a292"
14 ],
15 "urgency": 15.8
16}
What this means in normal English is that I'd made a website, but hadn't added the icons. It was getting late, so I made a note to remind myself to put the icons in later. If that note were just a 'todo.txt' file inside the folder with the website then I wouldn't remember to look, so it wouldn't get done.
It's not a very readable format, and pretty awful for searching through the text with text-search programs, but Taskwarrior could find whatever you wanted to know, so no problem there.
What's Great about Taskwarrior
Waiting
When you have a website, you need to renew the certificates. It takes two minutes. It's one command. But you don't need to do it now, you need to do it in a couple of months. I have seen businesses fail at this task. I've never let mine slip, because Taskwarrior will at that task to the list after a couple of months, before the time it's needed.
But not now. I don't want to look at 20 tasks that can't be done now every day. Taskwarrior will hide these until the right time, then gradually increase its position on the list as you approach the 'due' date.
Typing-Completions
Without this feature, nobody could use Taskwarrior. It lets you type full thoughts like 'show me tasks for this project which are tagged "css"', or 'what was that card-game thing I worked on last year?'.
Each command-word does not need more than three characters and if you make a
very long project name or massive tag (like +prognostication) then you just
type three characters and press the tab-key, and Taskwarrior completes the
word.
Unified Lists
You can pull a copy of tasks from Gitlab.com and similar places, which means
you have all the 'things to do' in one place, which is much nicer than having a
nagging feeling that you need to check a website, or sticking notification
emails in the email's /INBOX/Todo/ folder.
It doesn't cover every task-list-website in town, but it covers lots.
Stats
'Just a static website', I thought to myself, 'could be done in an hour, surely'. Everyone who works in IT suffers periods of debilitating optimism, where we think that you can build a castle in three days with sheer focus despite the fact that the dishes aren't done. Taskwarrior soothes the curse by reminding you that writing a little script to count how many files are in a folder took an hour last time, and making a static website took two weeks.
Problems
It Arrives with Tasks
When you update Taskwarrior, it comes with 'news'. You can read about it by
typing task news, and if you don't then Taskwarrior will remind you to do so.
This is a small task, but also a meta-meta-task. It's not something you actually need to do - it's something to do in order to keep up with the tool that keeps track of other tasks. All big projects (and collections of small projects) need a tool to keep track of them and the tool needs time to log and review tasks. But this is not that. It's not a meta-task. It's a meta-task about the tool. A meta-meta-task. The thin end of the wedge. The beginning of the end.
Calendars: Double or Nothing
Tasks have a 'scheduled' date (when you think it would be a good idea to actually do the thing), a 'due' date (when you really need to have done the thing) and an 'until' date (after which the task has been left so long it's no longer relevant, and will be automatically deleted with a note informing you of your failure).
Dates usually go in calendars, so how do you keep track of what you're doing on a given day? Taskwarrior does not work as a calendar, but it has dates. So you can either decide what things count as tasks and what counts as an appointment, then divide what you're doing between the two, or put them in both places, or simply ignore one.
I have mostly ignored my calendar for the last few years. This has not paid out well.
Synchronization
Taskwarrior has no straight-forward way to get the tasks onto your phone. It has many ways, but they all need a lot of setup and maintenance, and probably don't work (I never managed to set one up).
If anyone wants to look at or modify their to-do list on their phone they have just called for another Taskwarrior project, where step 1 is finding out how to do that, then writing down the other steps, and debugging the whole thing at a surprise-point in the future.
But surely computers work? Two computers running Linux should be able to share tasks, right? Alas, no. Taskwarrior wants all notes in a single file (not one file per project, or one per week). It must be one, and each task takes a line.
You might want to copy that file to another computer, back and forth, but that's a trap. If you take your laptop out somewhere without diligently turning it on, synchronizing, and then putting it back to sleep, then the laptop has an old list of tasks. Once you get back and try to copy changes to your PC, you have two different task lists, and have to manually sort out the mess.
Given the json structure above, you might almost get away with using git
and just telling the computer to automatically pick one or the other version
(who really cares when a task was 'last modified' when it's marked as
'complete'?). But the actual data looks like this:
1{"id":0,"description":"place new icons","due":"20241215T235959Z","end":"20241216T002703Z","entry":"20241215T215658Z","modified":"20241216T002703Z","project":"splint","status":"completed","uuid":"1d9d7cb5-67ea-41fd-bafc-795c3e74806e","tags":["web"],"depends":["e082f546-fef9-4a3d-b05e-e30d22e5a292"],"urgency":15.8}
Since the date shifts, that automatically shifts the 'urgency' rating, which means yet more conflicts even when the information hasn't changed.
An Engineered Collapse
Taskwarrior was very much 'complete' software, except for the complete lack of synchronization. Then along came 'Task champion' - a highly opinionated take on Taskwarrior, and how great it could be, if only it had a structured database backing it up. The code-changes were merged in at one point.
You might expect this to be in 'task news' before the scheduled date, but no -
you go to update and it all breaks. task news informs you that you should
already have exported tasks from the old version in order to import them.
Obviously a small database sounds great. One little table in a (an?) sqlite
database could synchronize between machines by adding a second table for
changes, and then swapping the change-elements so that a task marked as 'done'
in one would also become 'done' in another.
Let's look at the tables inside this new style of task-list.
| Table | Rows | Cols | keys |
|---|---|---|---|
operations |
113957 | 5 | id |
sqlite_sequence |
1 | 3 | |
sync_meta |
1 | 3 | key |
tasks |
9961 | 3 | uuid |
working_set |
341 | 3 | id |
version |
1 | 4 | singleton |
sqlite_stat1 |
5 | 4 | |
sqlite_stat4 |
33 | 7 |
What in the bleeding fuck is this?
I just wanted to remember to add icons to a website and have a run-down of how the projects went over the last month. This feels like too much for something that many people handle with some post-it notes.
But at least the data has a proper structure - it can has a 'project' column,
instead of being a bunch of unsightly json data. All well-structured and
easy to query.
| uuid | data |
|---|---|
| 57363268-f4c2-47a0> | {"entry":"1029083116","status":"pending","modified":"1080214120","description":"trans> |
| 3390c445-8418-4f1e> | {"status":"pending","modified":"2640814121","project":"ttrpgs","description":"transfe> |
| d4ddccf5-c87b-4f76> | {"modified":"3680234122","description":"connect gitea","entry":"1689143137","project"> |
| bc2f0d5a-5916-4291> | {"project":"hexbind","modified":"1679445877","description":"make list of rules for va> |
| ffc67fd1-895c-484a> | {"modified":"3678000282","dep_bc2f0d5a-5916-4291-9e6e-b338951c1cc5":"x","dep_8382df4e> |
| 7c11c83b-8ae8-4743> | {"entry":"4675969110","annotation_1675949134":"https://dungeonscrawl.com","status":"p> |
Again, what in the nine hells am I looking at? I don't work with databases,
but it looks like the json structure just got stuffed into a single row,
which feels about as natural as stuffing your hair in the kitchen sink's
drain-pipe. Instead of asking for task WHERE 'project' == 'website', the
program takes task 0001, spits out a json shoe-lace, parses each bit until
it arrives at the word "project":, then takes asks if the result equals
website. Once it continues through tasks 0002, then 0003, then all the
up to 9960, it finally creates rough estimate of how fast the project's
moving along.
The old version listed tasks for the day and gave project burn-downs before
your pinkie-finger lost contact with the return-key. The new sqlite version
gave my laptop an asthma attack for about five seconds before choking out an
answer. Others reported the same issue.
Obviously the people who made these changes cannot imagine humans who run command-line tools without paying €5,000 for a graphics card.
I should have jumped ship at this point.
Synchronization at Last
With the new database 'working', Taskwarrior could finally allow you to keep track of tasks on a PC and a laptop at the same time (and perhaps one day, a phone!). The manual provided two methods:
- A Google bucket.
- An Amazon bucket.
This left only one small problem:
- I don't want to put my tasks in a bucket.
- Even if I did, I don't have a bucket.
- This universally accessible command-line tool now demands payment to a US corporation.
- The open-source program assumes reliance on a proprietary back-end.
- Every task on your laptop will make the journey through your router, out of the city, under the sea, up to outer space, back down to a data-centre in San Francisco, and then back again to your PC, like Bugs Bunny escaping Elmer FUD.
- When you inform your PC that
buy big rice baghas been done, a stipend must be paid to an organization responsible for a bit of genocide before the laptop can also become aware of this fact. - By default, this new Taskwarrior nags you about not setting up synchronization, like a covert Google advert.
It also left some larger problems, if you want to go via San Francisco.
Alternatively, you might use a local database. This has none of the problems
above. It has only one problem; it cannot synchronize with another computer.
It can only create a backup. You can copy that backup to another computer,
then add some tasks and synchronize, and then copy it back, but if both
computers gain a change without being aware of each other then the local
database will shriek and smash into a thousand shards, never to be repaired
without diving through seven layers of sqlite and json.
At this point, I should have switched to writing directly on my arms with a pen, then smearing the ink on my office wall and reading them backwards. The synchronization would have been smoother and more reliable.
Alas, I mounted that local database over ssh as a fake file system and set both computers to synchronize every five minutes, but never the same minute (lest corruption take hold), and tried to not move between the two of them too quickly, while ensuring I never left the house with the laptop unless the laptop had been given proper time to wake and take in the latest updates.
Synchronization with Git, at Last
A recent update brought the feature every sensible command-line-lover assumed
it would have from the start: it could now synchronize using git.
To do this, one first has to set a password, in plain text, in the
configuration file. Where does the configuration file go? Hopefully not in a
git.
I wondered why anyone would bother to encrypt their tasks, since one can just
make a private git repository on literally any computer with ssh installed.
Then I remembered all the IT folk I've met who will happily make 10,000 line
changes to a git, push it to a Kubernetes config, ask Jeeves to resolve any
merge conflicts, but cannot set up their own git server because they
genuinely think that git is how VSCode communicates with Github.
Unfortunately, it would only 'synchronize' with the git repository, not with other computers. It still cannot handle two machines with tasks without breaking.
I wonder if it's worth my time to comb through all the software that the designers have touched so I can avoid it. Tragically, I have no task-list in which to place this excellent task.