I accidentially broke my atuin database by executing `atuin` from the
PR I was working on without setting these variables and had to manually
roll back my local database.
That shouldn't happen, so we set the database and record store path in
the devshell to something that does not overwrite our normal databases.
We also warn if these files already exist, because when entering the
devshell, a user might want to start from a clean slate here.
Signed-off-by: Matthias Beyer <mail@beyermatthias.de>
Now that local history is stored encrypted, new_key should not overwrite
an existing one. This may be frustrating, but will remove the risk of
Atuin generating a new key and the user losing their old one.
* add support for getting the total length of a store
* tidy up sync
* auto call init if history is ahead
* fix import order, key regen
* fix import order, key regen
* do not delete key when user deletes account
* message output
* remote init store command; this is now automatic
* should probs make that function return u64 at some point
1. Test that multiple users can be registered without clobbering each
other
2. Test that one user can change their password without it affecting the
other
I'd like to also test sync with multiple users, to ensure we never
accidentally leak data cross-users.
For a few reasons
1. This step is really, really slow. I don't think there's sufficient
value in a slow CI step to keep it
2. Whenever we add an integration test it needs to be added to the
ignore list. I want to keep friction on adding such tests as low as
is possible.
3. We already run tests in a bunch of places, so I don't think this is
needed
Ref: #1123
* Add basic xonsh support
* Add init xonsh command
* Add Xonsh install instructions in docs
* Add xonsh ctrl-R search
* update xonsh script and instructions
Summary of changes:
* Added duration to postcommand hook
* Switched main search operation to use `subproccess.run()` rather than running as an xonsh shell command - this a) allows us to capture stderr without needing a temporary file and b) avoids a weird broken-buffer state that results from running a fullscreen TUI and then programmatically editing the buffer
* Added support for immediately executing chosen command via `__atuin_accept__:` (like bash/zsh/fish)
* strip newline from command before sending to atuin
* Add basic xonsh support
* Add init xonsh command
* Add xonsh ctrl-R search
* Remove advanced-install guide (was accidentally re-added during rebase)
* Clean up
Xonsh doesn't import private functions into the local namespace when sourcing a file
* Add xonsh ro readme
* Respect ATUIN_NOBIND
* Format with black, and improve PEP8 compliance
* Add up search
* Format rust code
---------
Co-authored-by: Joseph Montanaro <jfmonty2@gmail.com>
* Add change-password command & support on server
* Add a test for password change
* review: run format
---------
Co-authored-by: Ellie Huxtable <ellie@elliehuxtable.com>
* feat: make history list format configurable
* Update atuin-client/config.toml
* review: run format
---------
Co-authored-by: Ellie Huxtable <ellie@elliehuxtable.com>
The current widget names for Zsh start with "_", which gives an
impression to users that those widgets are internal API and should not
be bound by the users. However, we actually instruct users to set up
custom keybindings by specifying them to bindkey.
In other shells, a separate namespace for widgets are not prepared, so
we want to prefix "_" to shell function names to tell the users that
these are not the commands that are supposed to be called from the
command line. However, the widget names are separated in their own
namespace in Zsh, so we do not have to isolate them by prefixing "_".
In fact, other frameworks such as `fzf` define widgets with names not
starting with "_".
In this patch, we update the widget names to have the form "atuin-*".
The old widget names that existed in the release version <= 17.2.1 are
left for compatibility.
A search feature in the vi-normal mode of shells can be called by "/".
To make the shell-integration keybindings consistent with the existing
shell keybindings, we add the keybinding to "/" in this patch.
This patch removes the Ctrl-r binding in the "vicmd" keymap in Zsh.
The key Ctrl-r is used for the `redo` operation in the vi-normal mode
by default. We would like to avoid overwriting an existing keybinding
with a totally different feature.
* chore: add feature to allow always disable check update
In the packaging rules of some distributions, the software's self-update check needs to be permanently turned off
This commit will make it easier for these users to
* fix: formatting
---------
Co-authored-by: Ellie Huxtable <ellie@elliehuxtable.com>
The runtime formatting used to list history commands can fail in its
Display implementation and the error is not propagated through to the
write error it causes. Instead, the error is cached in the FormatArgs
being printed. It's a bit clumsy but by checking if there's a cached
error, we can get a more useful error for the user. eg,
atuin search cargo --format '{invalid}'
gives
history output failed with: The requested key "invalid" is unknown
instead of
history output failed with: formatter error
* fix(stats): Don't bail/error if no command is found
An empty history shouldn't be source of error when printing stats.
* fix(stats): Improve help message a bit.
It wasn't clear what the period format could be.
* fix(bash): strip control chars generated by \[\] in PS1 with bash-preexec
Fixes https://github.com/atuinsh/atuin/issues/1617
* perf(bash): count newlines using Bash built-in features
When piping the output of `atuin history list` to a file, it makes more
sense for the literal commands to be written rather than the escaped
ones that would be printed to the terminal.
An issue with the old sync was that if there was _one_ record encrypted
with a different key, sync would stop. You'd need to delete your account
and start from scratch. This sucked.
This change means we will carry on, and try to encrypt and build with as
much of the history as we are able to decrypt.
This is possible because we can quite happily store data on disk that we
cannot decrypt. The old store couldn't do this.
In future, we might consider a keyring containing multiple keys.
Related: https://forum.atuin.sh/t/search-ignoring-commands/74/5?u=ellie
When a user ran a duplicated command, but in another session, it was
removed by filters. This is because the subquery that was once used did
not have the same filters applied as the main query.
Instead of messing with subqueries, `group by` instead. This aligns with
the search() function
* feat(search): make cursor style configurable
The vim mode of the interactive Atuin search changes the cursor style
on a mode change, but the current implementation has the following
issues.
* The terminal's cursor style set by the Atuin search remains after
Atuin exits. This causes an inconsistency with the shell's setting
for the cursor style.
* Also, the cursor style for each keymap mode is currently hardcoded
in the source code, which is not necessarily consistent with the
user's cursor-style setting in the shell.
* Since the current implementation does not set the cursor style for
the initial keymap mode but only sets the cursor style when the
keymap mode is changed, it also causes inconsistency in the cursor
style and the actual keymap when the shell's keymap and Atuin's
initial keymap mode are different.
This patch solves those issues by introducing an opt-in configuration
variable `keymap_cursor`. By default, the vim mode does not change
the cursor style because there is no way to automatically determine
the cursor style consistent with the shell settings. We enable the
feature only when the user specifies the preferred cursor style in
each mode in their config. Also, the cursor style is set on the
startup of the Atuin search (based on the initial keymap mode) and is
reset on the termination of the Atuin search (based on the shell's
keymap mode that started the Atuin search).
* chore(settings): remove dependency on crossterm
* fix: add acquire timeout to sqlite database connection
This should fix#1503
I wasn't able to trigger enough IO pressure for the SQL connection to be
a problem.
This adds `local_timeout` to the client config. This is a float, and
represents the number of seconds (units in line with the other timeouts,
though those are ints). Users may well want to reduce this if they
regularly have issues, but by default I think 2s is fine and avoids a
non-responsive system in bad situations.
* tests
When printing the history list with either the session or cwd filter
enabled, use to same query method as without either to ensure that the
other options (hide deleted entries etc) are respected.
* Print literal control characters to non terminals
Previous 'fix' to prevent control sequences being interpreted when they
shouldn't have been also prevented them being used when they should have
been. This checks if the output is to a terminal (where control
sequences shouldn't be interpreted) before escaping control characters.
* Update atuin/src/command/client/search.rs
Co-authored-by: Ellie Huxtable <ellie@elliehuxtable.com>
---------
Co-authored-by: Ellie Huxtable <ellie@elliehuxtable.com>
If a previous command in the history contained a literal control
character (eg via Ctrl-v, Ctrl-[), when the command was printed, the
control character was printed and whatever control sequence it was part
of was interpreted by the terminal. For instance, if a command contained
the SGR sequence `^[[31m`, all subsequent output from `atuin history
list` would be in red.
Slightly less of a problem, control characters would also not appear in
the interactive search widget although they would be printed when
selected. This meant `echo '^[[31foo'` would appear as `echo '[31foo'`.
When the entry was selected, the same problem as before would occur and,
for the example above, `echo 'foo'` would be printed with 'foo' in red.
When copied, this command would not behave the same as the original as
it would be missing the control sequence.
This adds an extension trait to add a method to anything that behaves
like a string to escape ascii control characters and return a string
that can be printed safely. This string can then be copied and run
directly without having to add the control characters back.
* feat: add history rebuild
This adds a function that will
1. List all history from the store
2. Segment by create/delete
3. Insert all creates into the database
4. Delete all deleted
This replaces the old history sync.
Presently it's incomplete. There is no incremental rebuild, it can only
do the entire thing at once.
This is ran by `atuin store rebuild history`
* fix tests
* add incremental sync
* add auto sync
Imagine the scenario where a sync fails. The function touched here will
never save the sync time.
That means that external to this function, the sync never happened. The
next command will try and start one immediately.
The happy scenario is that this succeeds. The unhappy scenario is that
this fails. Fails, and isn't saved, so we try again...
Should add proper backoff but this is a good start.
* Fix fish integration on older fishes
- On fish 3.3 the bash-style `$(...)` doesn't work, we should use `(...)`
- Also quoting `"(moo)"` on older fishes gives a literal `moo`
- The result of a `(subcommand)` is a single token, so no need to quote it, anyway
Tested by making the change, executing `cargo run -- init fish --disable-up-arrow` and then
executing that shell script on a system with fish 3.3, as well as 3.6 and observing both still work.
Fixes#1562.
* Alternative way of fixing the problem. Compatible with 3.3 and 3.4+
