This is instead of a single big string with chunks separated by ":".
":" is used in RFC 3339 date-time strings (like "2025-07-16 21:23:44
-06:00"), so it's inconvenient to reserve ":" to be the chunk separator.
I'm not super wedded to this new Vec<String> way of doing the filter,
but it seems fine and convenient for now.
Instead of a two-step process handled by the application
(`Issue::new_comment()` and `Comment::set_description()` or
`Comment::edit_description()`), make a simpler-to-use single-step
`Issue::add_comment()`.
Move the implementation details from Issue to Comment.
Better log message when adding a comment.
This starts cleaning up the Issue API.
* Start separating the public API from the internal API.
* Make `Issue::new()` and `Issue::edit_description()` better behaved,
simpler, reduce code duplication, and also produce better git log
messages.
* Update `ent` to call the changed `new()` function.
* Add some comments documenting the Issue API.
* `ent new` and `ent edit` now use the editor specified by the EDITOR
environment variable, if any. Defaults to `vi` if unspecified.
This makes each command in `handle_command()` handle its own issues
database. The commands that only need read-only access to the issues
make a detached worktree, while the commands that need read-write access
make a normal worktree with the `entomologist-data` branch checked out.
This lets us run any number of read-only operations (like `ent list`,
`ent show`, etc), even when a long-lived read-write operation (like
`ent new` or `ent edit`) is running.
Fixes `ent 317ea8ccac1d414cde55771321bdec30`.
right-speech-bubble is not part of the recommended unicode
emoji presentation set, but left-speech-bubble is and therefore, it's
more commonly supported
I'm not sure about the filter format...
There are two independent filters: "state" and "assignee". The "state"
filter defaults to including issues whose state is InProgress, Blocked,
Backlog, or New. The "assignee" filter defaults to including all issues,
assigned or not.
The two filters can be independently overridden by the `ent list
FILTER` command. FILTER is a string containing chunks separated by
":", like the PATH environment variable. Each chunk is of the form
"name=value[,value...]". "name" can be either "state" or "assignee".
The "value" arguments to the "state" filter must be one of the valid
states, or it's a parse error.
The "value" arguments to the "assignee" filter are used to
string-compare against the issues "assignee" field, exact matches are
accepted and everything else is rejected. A special assignee filter of
the empty string matches issues that don't have an assignee.
Some examples:
* `ent list` shows issues in the states listed above, and don't filter
based on assignee at all.
* `ent list assignee=seb` shows issues in the states listed above,
but only if the assignee is "seb".
* `ent list assignee=seb,` shows issues in the states listed above,
but only if the assignee is "seb" or if there is no assignee.
* `ent list state=done` shows all issues in the Done state.
* `ent list state=done:assignee=seb` shows issues in the Done state that
are assigned to "seb".
This commit makes a couple of changes:
- `ent show ISSUE` now displays the Issue's Comments in chronological
order
- the Comment struct now includes a timestamp, which is the Author Time
of the oldest commit that touches the comment's directory
- the Issue struct now stores its Comments in a sorted Vec, not in
a HashMap
- The Comment's uuid moved into the Comment struct itself, instead of
being the key in the Issue's HashMap of Comments
In a worktree with the `entomologist-data` branch checked out in it:
1. `git fetch REMOTE`
2. `git merge REMOTE/BRANCH`
3. `git push REMOTE BRANCH`
Pretty straight-forward. If anything goes wrong we error out and ask
the human to help.
