After _completion_loader is called, instead of loading the entire
completion.bash file, just restore the fzf completion for the current
command. `_fzf_orig_completion_$cmd` is only set if _completion_loader
actually changed the completion options to avoid infinite loop.
Close#1170
When `ps` is aliased for something uncommon, like `alias ps=grc ps` which colorizes ps output, the output of `ps` can be unexpected and/or undesired.
This change makes ps to be always executed as command, even if it's aliased.
Handles records like "[20.20.7.168]:9722 ssh-rsa ..."
This is a standard format for servers running on custom port according to http://man.openbsd.org/sshd.8#SSH_KNOWN_HOSTS_FILE_FORMAT
A hostname or address may optionally be enclosed within ‘[’ and ‘]’
brackets then followed by ‘:’ and a non-standard port number.
* Don't do shell quoting for weird chars
This would prevent tabs from being escaped as `$'\t'` (definitely not what I would want to see as initial value in the search).
* Do different escape.
This also modifies <C-t> behaviour.
The longest file path in the input is used as root directory for `find`
command. The remainder of the input is passed to fzf's --query as a
initial search parameters.
If "." is given as the argument to begin <C-t> completion, the leading
"." is not correctly removed. In general, if user selects a fzf
completion, the current token should be "consumed".
CTRL-R binding used to start with --no-sort to list the matched commands
in chronological order. However, it has been a constant source of
confusion. Let's enable it by default from now on. The sorted result
shouldn't be too confusing as we use --tiebreak=index.
_fzf_completion_loaded is no longer checked. This change increases the
load time by a few milliseconds, but I can't think of a better way to
handle the issue.
Close#783.
Replace the "temp file" workaround with the "read" function: it's
simpler and faster.
Use proper escaping, remove the custom function.
The "file" widget uses last token as root for the "find" command.
This replaces the equivalent of '**' completion in bash/zsh.
The "$dir" non-expanded variable can be used in FZF_CTRL_T_COMMAND to
set the root.
Handle uppercase letters in program names. This also deals with `-` and
`.`, both of which are quite common in program names, e.g., `xdg-open`
and `foo.sh`.
The command substitution and following word splitting to determine the default
zle widget for ^I formerly only works if the IFS parameter contains a space. Now
it specifically splits at spaces, regardless of IFS.
zle automatically calls zle-line-init when it starts to read a new line. Many
Zsh setups use this hook to set the terminal into application mode, since this
will then allow defining keybinds based on the $terminfo variable (the escape
codes in said variable are only valid in application mode).
However, fzf resets the terminal into raw mode, rendering $terminfo values
invalid once the widget has finished. Accordingly, keyboard bindings defined
via $terminfo won’t work anymore.
This fixes the issue by calling zle-line-init when widgets finish. Care is taken
to not call this widget when it is undefined.
Fixes#279
This allows to have a custom widget like the following, which would
additionally accept the line, but only in case of entries being
selected:
fzf-file-widget-with-accept() {
zle fzf-file-widget
if [[ "$?" == 0 ]] && (( $#BUFFER )); then
zle accept-line
fi
}
zle -N fzf-file-widget-with-accept
bindkey '\e^T' fzf-file-widget-with-accept
With this `<C-a>t` will launch fzf, and simulate the pressing of "Enter"
afterwards.
Faster startup. Use internal bash globbing instead of external grep binary (adapter from Gentoo's `/etc/bash/bashrc` TERM checking). Insignificant on Linux, but on Cygwin this cuts startup time by 40 ms on my Core i7 laptop.
Summary:
Fix adapted from [@adamheins: fzf, vi-mode, and fixing delays][1].
[1]: https://adamheins.com/blog/fzf-vi-mode-and-fixing-delays
The basic problem is that
fzf presses <Esc> to enter vi-movement-mode
(as opposed to insert mode)
and then presses a bunch of keys to set up the buffer.
But the <Esc> keypress is also the prefix for a bunch of other commands,
so Bash will dutifully wait an excruciating half-second
before actually executing this command.
Instead, we bind <C-x><C-a>, which is unused by default
and seems reasonably unlikely to be custom-bound,
to be another way to enter vi-movement-mode;
this binding is unambiguous, so fzf can use it without delay.
This change was made by just `:s/\\e/\\C-x\\C-a/gc`
in the relevant section,
after adding the actual binding and comment at the top.
Notes:
- You can now override _fzf_compgen_path and _fzf_compgen_dir functions
to use custom commands such as ag instead of find for listing
completion candidates.
- The first argument is the base path to start traversal
- Removed file-only completion in bash, i.e. _fzf_file_completion.
Maintaining a list of commands that only expect files, not
directories, is cumbersome (there are too many) and error-prone.
TBD:
- Added $FZF_COMPLETION_DIR_COMMANDS to customize the list of commands
which use directory-only completion. The default is "cd pushd rmdir".
Not sure if it's the best approach to address the requirement, I'll
leave it as an undocumented feature.
Related: #406 (@thomcom), #456 (@frizinak)
This seems like a bug of fish, but sometimes when you select an item
fish complains:
"insertion mode switches can not be used when not in insertion mode"
This only happens when using tmux pane. Injecting a dummy command
somehow fixes the issue.
Instead of choosing one at initialization, choose the correct one
when it's actually called, so that the behavior is correct even after
resizing.
Bonus fixes for tmux with bash 4+:
- No extra space when cancelling CTRL-T.
- Fix cursor position problem in vi mode.
In the case that fzf-tmux returns a user-selected result but with a
non-zero exit status (which can happen if a function inside $PS1 returns
non-zero) this allows CTRL-R to continue working as expected.
Addresses #203 (Tranquility's comment)
* _fzf_complete is the helper function for custom completion
* _fzf_complete FZF_OPTS ARGS
* Reads the output of the source command instead of the command string
* In zsh, you can use pipe to feed the data into the function, but
it's not possible in bash as by doing so COMPREPLY is set from the
subshell and thus nullified
* Change the naming convention for consistency:
* _fzf_complete_COMMAND
e.g.
# pass completion suggested by @d4ndo (#362)
_fzf_complete_pass() {
_fzf_complete '+m' "$@" < <(
local pwdir=${PASSWORD_STORE_DIR-~/.password-store/}
local stringsize="${#pwdir}"
find "$pwdir" -name "*.gpg" -print |
cut -c "$((stringsize + 1))"- |
sed -e 's/\(.*\)\.gpg/\1/'
)
}
# Only in bash
complete -F _fzf_complete_pass -o default -o bashdefault pass
While in bash you can externally register custom completion functions
using `complete` command, it was not possible to do so in zsh without
changing completion.zsh as the name of the supported commands are
hard-coded within the code (See #362). With this commit, fzf-completion
of zsh will first look if `_fzf_COMMAND_completion` exists and calls the
function, so one can externally define completion functions for specific
commands.
This commit also tries to make the interface of (yet undocumented)
_fzf_list_completion helper function consistent across bash and zsh.
So the following code works both on bash and zsh.
_fzf_pass_completion() {
local pwdir=${PASSWORD_STORE_DIR-~/.password-store/}
local stringsize="${#pwdir}"
let "stringsize+=1"
_fzf_list_completion '+m' "$@" << "EOF"
find "$pwdir" -name "*.gpg" -print | cut -c "$stringsize"- | sed -e 's/\(.*\)\.gpg/\1/'
EOF
}
# Only on bash
complete -F _fzf_pass_completion -o default -o bashdefault pass
Note that the suggested convention and the interface are not yet final
and subject to change.
/cc @d4ndo
