While working on making the history command support case-sensitive and
insensitive searches I noticed that entering "all" when interactively
deleting history entries resulted in an error. That's because the
history builtin currently only supports `--exact` so we need to loop
over the matching entries and delete them one at a time.
Fixes#3448
My previous change to avoid creating a *.pyc file when running
create_manpage_completions.py was wrong because I put the
`sys.dont_write_bytecode = True` on the wrong line. Rather than simply
move that statement make the simpler, cleaner, fix that removes the need
for `eval` where that program is invoked.
The Linux kernel only splits on the first whitespace in the shebang line
(unlike BSD which splits on all whitespace). Which means there can be
only one argument after the path to the program.
Producing man pages is done infrequently (basically just at `make test`
and `make install`) so there isn't any point in writing compiled
byte-code versions of the python modules.
After implementing `builtin fish_realpath` it was noticed that it did
not behave like GNU `realpath` without options. Which is super annoying
since that was the whole point of implementing the command. Major
failure on my part since I wrote the unit tests to match the behavior of
the existing `wrealpath()` function that I simply exposed as a builtin
command. Rather than actually verifying it behaved in a manner
compatible with GNU realpath.
Also, while the decision to call the builtin `fish_realpath` seemed to
make sense at the time of the original commit further reflection has
shown that to be a silly, idiosyncratic, thing to have done. So rename
it to simply `realpath`.
Fixes 3400
This adds a flag to the `history search` command to limit the number of
matching entries to the first "n". The default is unlimited. This is
mostly useful in conjunction with aliases (i.e., functions) that are
intended to report the "n" most recent matching history entries without
piping the result through the user's pager.
Fixes#3244
Don't wrap fish_indent at all if the version in $PATH matches
$FISH_VERSION.
When we do wrap it, resolve the path once, and use that via alias
machinery instead of doing an eval each time.
In both cases, `type fish_indent` can tell us what it's actually going
to do now.
clarity aside, it's faster if we only eval the one time.
eval is not only evil, but slow.
> for h in $history[1..100]; echo $h | fish_indent --no-indent; end
before: CMD_DURATION = 1005
if fish_indent is kosher in PATH: 549
if not, using alias: 687
these modern terminals both compose a nicer title if we don't try to provide a custom one (no path in title twice, "fish" in title twice) - and the user can configure which components they'd like in their terminal inside the terminal preferences.
Also make test "$VTE_VERSION" -ge .. work once I commit `test` strtoi
fix - the trick is to add a zero before it so the numeric comparison
works even if it's empty.
Fixes#107
This deprecates the use of long options for history sub-commands (e.g.,
`history --delete`) in favor of proper sub-commands (e.g., `history
delete`). It also eliminates the short options for those sub-commands.
Also change option processing to allow options anywhere on the command
line to match how the vast majority of fish builtins handle flags.
Replace --with-time with --show-time.
Fixes#3367
* Use 'grealpath' if installed for realpath fallback
See discussion in #3370
* fish_realpath: filter out dangerous options
Per feedback do not use aliases to declare wrapped functions.
chown completion chown currently uses cat /etc/group to fetch the list of group names. In Cygwin there's no /etc/group file any more (user and group names are fetched directly from the OS), so when a user tries to tab-complete the group name they get an error message:
ASchulma@LZ77E1AASCHULMA ~/d/fish> chown ASchulma🐱 /etc/group: No such file or directory
This change fixes that by using getent group (via __fish_complete_groups) by preference to get the group names, and falling back to /etc/group. This is more portable.
This has the same name and path as ubuntu's, but takes less arguments.
So we need to actually find if the distro thinks it is suse, and then
use it.
Fixes#3366.
Adds a color reset thing, to ensure fish tries to use hard colors during
testing.
Also, work on a discrepancy (not introduced by my changes, afaik) when
with some combinations of color settings, and usage of --bold, caused super
flakey color paninting in the pager. Downwards movements that trigger
scrolling vs. upwards movement in the pager would only apply bold to
selections when moving upwards. The bold state of the command completions in
the pager was flipping flops on and off, depending on if there is a description
on the preceding line.
Implement a lame fix by reseting the color to normal and applying a
different style on the rightmost ')' which seems to be what was influencing it.
Makes fish use terminfo for coloring the newline glich char.
The previous solution would not erase the previous bindings if
fish_vi_key_bindings was called with a mode argument. So if the user
switched to vi with a different initial mode, they'd keep their previous
bindings also.
Supersedes e89057b.
Some of these were defined in the shared bindings, some (like \cy yank)
were just literally duplicate in the same files.
This should _not_ change anything. In particular this does not remove
hardcoding of sequences (because terminfo might be wrong or the term
might need smkx).
Found with
```
function bind
set -l binds (builtin bind)
builtin bind $argv
set -l newbinds (builtin bind)
if set -q argv[1]; and not test "$argv[1]" = "--erase"
if test "$binds" = "$newbinds"
echo "Duplicate: " (string escape -- $argv)
end
end
end
```
The vi-bindings function would unconditionally erase all bindings,
making it impossible to call it last. This would disable the
mode-indicator (and in future also the cursor).
Make it so any argument to fish_vi_key_bindings stops it from erasing
bindings.
It would also be possible to demand an argument to erase (or to erase as
a separate step). but the usual case seems to be _switching_ to a set of bindings.
This didn't work on platforms where tput exists but can never accept
terminfo names. This includes the current versions of FreeBSD - it
used to do both, now it doesn't. So, fall back to the old termcap names
by (tput smso; or tput so). Add check for the tput program before we
even try.
The extra things `eval` does are all for code that runs
interactively. Because we just define a function, we don't need it.
This improves alias' performance by about 20-25% (0.783608s to 0.585585s
on about 500 aliases) and avoids triggering #3345.
This can be prohibitively slow on large repositories (minutes!).
While regrettable, no user is going to like waiting that long.
Work towards #3342, rerun of #3230.
Many thanks to @gladhorn for the idea!
Offering auto completion for existing commits is great, but on big
repositories, it suddenly becomes really slow, even with fast hard
disks, since each commit is read and then a line processed for it.
Instead limit to the last 500 commits (arbitrary number) which still
feels fast. Going back further in history can easily and more reasonably
done with git log etc.
* completions/p4.fish
* Updated per comments + added p4 clients
* p4 completions: integ, opened, reopen. "default" CL support.
* Perforce RCS -> SCM
* p4 reopen: list opened files
* Fixed per review, added -d for all functions
Fixed per comments in review by @faho,
Added -d for all functions,
Renamed ”subcommand" term to “command” (so there’s probably diff noise)
* p4 completions with submit list of files
* p4 completions for submit: lists open files
Implementing the --shadow-builtin flag has proven to be highly controversial.
Revert the introduction of that flag to the `function` command. If someone
shoots themselves in the foot by redefining a builtin as a function that's
their problem and not our responsibility to protect them from doing so.
Fixes#3319
It's not ideal since we can't get the real result so we just assume it's
"0". That triggers the easier path, which still might display the wrong
thing, but we have to pick something.
Possible fix for #3321.
* fixes broken completion of screen on osx, test on ubuntu and mac with fish 2.3.1
* replaces sed, __fish_sgrep with fish builtin string
* add completion for `screen -x`
* adjust format (e.g. 12345.socket\t01/01/16 09:55:00 Detached)
* Fix brew completion for `brew install`
* Using `brew search` rather than `brew --repository`
- Homebrew migrated the directory holding their Formulas into Taps, breaking fish's completions.
- New method to find all Homebrew-core Formulas
- Compatible with old versions of Homebrew and more future proof
* Replace fixed path to search formula with `brew --repository`
* Replace `sed` with builtin `string replace`
The recent change to reconcile the history builtin command and function
broke an undocumented behavior of `history --delete`. This change
reinstates that behavior. It also adds an explicit `--exact` search mode
for the `--search` and `--delete` subcommands.
Fixes#3270
Only on the OS X travis build.
I can't reproduce it but I figure it's something to do
with test -e vs test -x or the echo -n in command substitution.
Oops.
This was erroneously omitted from the previous commit.
Now backspace in insert mode does backward-delete-char, in default mode
backward-char (i.e. no deleting, just moving). This is consistent with vim.
This undoes the inheritance since it shared too much.
The idea here is to share bindings that aren't something the editors we're inspired by do - there's no "execute" in vi.
The basic editing and moving bindings are now vi-style in vi-mode and emacs-style in default mode.
In particular, Screen and emacs' "ansi-term" behave like neovim in that
they just ignore the sequences, which leads to the terminal rendering
default color (most of the time white) instead.
Only one file belonging to fish-shell had DOS/bogus line endings,
with `git add' picking up changes after updating .gitattributes:
hostname.fish.
Unsurprisingly, it has code to support cygwin and was likely
worked on by a user on a Windows machine. This will help
such cases in the future.
Also, in pcre2-10.21/, there was RunTest.bat which was (correctly)
CRLF formatted. We don't use this batch script at all, so rather
than LF it or add an exception, blast it away like the other pcre2
files omitted from the repo.
A common problem for users is that fish doesn't get a locale. This often
happens if systemd is used with getty and fish as login shell.
Fixes#277
Note that I (@krader) made editorial changes before merging this. For
example, running `make style` and otherwise changing long statements to a
series of shorter statements. So if there are any problems it is possible
I introduced them.
Fish assumed that it could use tparm to emit escapes to set colors
as long as the color was under 16 or max_colors from terminfo was 256::
if (idx < 16 || term256_support_is_native()) {
// Use tparm to emit color escape
writembs(tparm(todo, idx);
If a terminal has max_colors = 8, here is what happenened, except
inside fish:
> env TERM=xterm tput setaf 7 | xxd
00000000: 1b5b 3337 6d .[37m
> env TERM=xterm tput setaf 9 | xxd
00000000: 1b5b 3338 6d .[39m
The first escape is good, that second escape is not valid.
Bright colors should start at \e[90m:
> env TERM=xterm-16color tput setaf 9 | xxd
00000000: 1b5b 3931 6d .[91m
This is what caused "white" not to work in #3176 in Terminal.app, and
obviously isn't good for real low-color terminals either.
So we replace the term256_support_is_native(), which just checked if
max_colors is 256 or not, with a function that takes an argument and
checks terminfo for that to see if tparm can handle it. We only use this
test, because otherwise, tparm should be expected to output garbage:
/// Returns true if we think tparm can handle outputting a color index
static bool term_supports_color_natively(unsigned int c) { return max_colors >= c; }
...
if (term_supports_color_natively(idx) {
And if terminfo can't do it, the "forced" escapes no longer use the fancy
format when handling colors under 16, as this is not going to be compatible with
low color terminals. The code before used:
else {
char buff[16] = "";
snprintf(buff, sizeof buff, "\x1b[%d;5;%dm", is_fg ? 38 : 48, idx);
I added an intermediate format for colors 0-15:
else {
// We are attempting to bypass the term here. Generate the ANSI escape sequence ourself.
char buff[16] = "";
if (idx < 16) {
snprintf(buff, sizeof buff, "\x1b[%dm", ((idx > 7) ? 82 : 30) + idx + !is_fg * 10);
} else {
snprintf(buff, sizeof buff, "\x1b[%d;5;%dm", is_fg ? 38 : 48, idx);
}
Restores harmony to white, brwhite, brblack, black color names.
We don't want "white" to refer to color color #16, but to the
standard color #8. #16 is "brwhite".
Move comments from output.h to output.cpp
Nuke the config.fish set_color hack for linux VTs.
Sync up our various incomplete color lists and fix all color values.
Colors 0-8 are assumed to be brights - e.g. red was FF0000. Perplexing!
Using this table:
<http://www.calmar.ws/vim/256-xterm-24bit-rgb-color-chart.html>
Fixes#3176
This is a regression introduced by 834ebef53c
Bolster with a check for only login sessions too -- hopefully makes it
less annooying on subshells in general.
Fixes#3261
This fixes several problems with how the builtin `history` command handles
arguments. It now complains and refuses to do anything if the user specifies
incompatible actions (e.g., `--search` and `--clear`). It also fixes a
regression introduced by previous changes with regard to invocations that
don't explicitly specify `--search` or a search term.
Enhances the history man page to clarify the behavior of various options.
This change is already far larger than I like so unit tests will be added
in a separate commit.
Fixes#3224.
Note: This fixes only a couple problems with the interactive `history
--delete` command in the `history` function. The main problem will be
dealt with via issue #31.
* git completion: add mergetool
The list of tools is stole from the bash completion file that comes with
git.
* git completion: complete files with merget conflict for mergetool
We don't seem to mention in the documentation that we were forcing
-t for all interactive uses. If we want to do that we should apply
that in the builtin.
history.fish reimplementing every option and doing things kind of
differently is a real pain and it's not clear if the docs are
referring to the or the wrapper script or both.
Problem with Type2ManParser
before:
complete -c xcode-select -s h -l help --description 'Prints the usage
message. UNINDENT NDENT 0. 0.'
after:
complete -c xcode-select -s h -l help --description 'Prints the
usage message.'
Don't truncate long lines with " [See Man Page]" suffix - use the
reclaimed 15 characters for more-useful usage info.
Improve the --verbose output with:
- spacing fixes
- diagnostics related to input print repr()/quoted as %r to be less
confusing.
- get rid of stupid name() and use type()/__class__.__name__,
- Always use new-style (new as in post python 2.2) classes so this
behaves the same whether we run in python 2 or 3.
- Properly convert left-quotes and right-quotes to that character in
deroff.py
Someone running fish in an unusual locale reported that an `assert()` was
firing when they typed `pkill c`. I traced it to two bugs. First, the
__fish_make_completion_signals command was producing a weird result. Second,
the builtin `complete` command wasn't adequately verifying its arguments.
Fixes#3129
Instead of just using Courier New across the board, have the
browser try several likely available fonts before defaulting
to the system's "monospace".
Thanks @MarkGriffiths
Fixes#2924
Doesn't colorize if output is redirected.
This is "fun" and indenting happens to make most of the included
functions display more narrow and fit better into a terminal window.
Doesn't colorize if output is redirected.
This is "fun" and indenting happens to make most of the included
functions display more narrow and fit better into a terminal window.
The previous commit to add a `--with-timestamp` flag to the `history` command
caused me to notice the history function didn't recognize the new long option.
Neither did it recognize the short options for the builtin command. This
change fixes both of those issues.
Improves experience during upgrades, accidentally running
an old fish with a new environment. No errors just from
printing a prompt. Fixes#3057.
Print helpful notice also when launching mismatched fish.
Autoloadable string.fish -- only create function if not builtin.
Completion throws and error about the command `__fish_contains_opts` beings unknown. It seems to be a simple typo, as all other completions use `__fish_contains_opt`
https://www.freebsd.org/cgi/man.cgi?query=ssh_config&sektion=5
1. It is possible to add multiple whitespace characters between the keyword (i.e. Host) and the argument(s).
2. It is allowed to have a single = and whitespace between the keyword and the argument(s).
3. It is possible to add multiple host names under a single Host directive by spacing the names apart.
1. and 3. are actual conventions that we use in our team, and I couldn't get auto-complete working for fish without this modification.
Modification explained:
a. The space between Host(?:name)? and the \w.* was replaced by (?:\s+|\s*=\s*) to match any sequence of whitespace characters, or optional whitespaces with a single =, per spec.
b. Result of first replacement is piped through another string replace to switch duplicate whitespace characters to a single space, and then piped to be split by that space. This allows specifying several aliases or host names in a single Host/Hostname definition, also per spec.
https://www.freebsd.org/cgi/man.cgi?query=ssh_config&sektion=5
1. It is possible to add multiple whitespace characters between the keyword (i.e. Host) and the argument(s).
2. It is allowed to have a single = and whitespace between the keyword and the argument(s).
3. It is possible to add multiple host names under a single Host directive by spacing the names apart.
1. and 3. are actual conventions that we use in our team, and I couldn't get auto-complete working for fish without this modification.
Modification explained:
a. The space between Host(?:name)? and the \w.* was replaced by (?:\s+|\s*=\s*) to match any sequence of whitespace characters, or optional whitespaces with a single =, per spec.
b. Result of first replacement is piped through another string replace to switch duplicate whitespace characters to a single space, and then piped to be split by that space. This allows specifying several aliases or host names in a single Host/Hostname definition, also per spec.
This was causing issues launching fish_config on OS X if fish.app is
renamed to contain a space (noted, but likely not the actual problem,
in issue #3140)
I mixed things up with `netctl` somehow. Since the two are quite
different they do not have the same function, they should not have
the same completions.
I also find that I would be smarter to only display the relevent
profiles given what we want to do. If we want to disable a profile
we should only complete with enabled profile for completion for
instance. I don't know if the implemention is nice enough however.
This makes fish_mode_prompt rely on $fish_key_bindings instead.
fish_bind_mode is also set in default mode (only always "default"), so
it can't be used as the indicator.
Closes#3067.
(cherry picked from commit 8ab980b793)
Completion throws and error about the command `__fish_contains_opts` beings unknown. It seems to be a simple typo, as all other completions use `__fish_contains_opt`
This speeds up the common case when IO is slow, e.g. when used with
sshfs.
We only use the short sha for figuring out whether the state is
valid (for which a long sha should also work) and for display when HEAD
is detached (I think that's the correct git-ism).
Working towards #3083.
Move to `string match` syntax from `grep` caused test to see if the Atom Package Manager is installed to always fail. This appears to fix the issue (tested on fish 2.3.0 with apm 1.6.0).
This potentially leads to an unusable session (when fish_key_bindings is
set in config.fish to a value without corresponding function), so we
should take care.
* Add missing color definitions to __fish_init_1_50_0 reset.
The values where determined by inspecting the values of:
* fish_color_end
* fish_color_user
* fish_color_host
after resetting the color theme via fish_config.
* Add documentation for fish_color_user and fish_color_host.
(cherry picked from commit 08c29727e0)
This is quite ugly because the syntax is ugly, the documentation both
under- and overspecified at the same time (a BNF that isn't...) and it
has a lot of functionality.
But the completion works half-decent for `ip address`, so let's ship it.
Instead of just using Courier New across the board, have the
browser try several likely available fonts before defaulting
to the system's "monospace".
Thanks @MarkGriffiths
Fixes#2924
It's currently too easy for someone to bork their shell by doing something
like `function test; return 0; end`. That's obviously a silly, contrived,
example but the point is that novice users who learn about functions are
prone to do something like that without realizing it will bork the shell. Even
expert users who know about the `test` builtin might forget that, say, `pwd`
is a builtin.
This change adds a `--shadow-builtin` flag that must be specified to
indicate you know what you're doing.
Fixes#3000
I'm going to modify these functions as part of dealing with issue #3000
and don't want those changes to be masked by running the files through
`make style`.
