Insert mode completions are accepted by typing any key. For example,
if there is a completion "somefunction()", then typing
some<c-n>;
will insert
somefunction();
and then the InsertCompletionHide hook will fire. The hook parameter
is a range that contains the entire thing: the actual completion plus
the trailing semicolon that closed the completion menu.
The [original motivation] for the hook parameter was to support
removing text inserted by completion, so we can apply text edits
or expand snippets instead. One problem is that we don't want to
remove the semicolon. Another problem came up in a discussion
about [snippets]: let's say we have a snippet "add" that expands to
add(?, ?)
where ? are placeholders. After snippet expansion the cursor replaces
the first placeholder. If I type "ad<c-n>1" I expect to get "add(1, ?)".
If the InsertCompletionHide hook only runs after processing the "1"
keystroke, this is not possible without evil hacks.
Fix these problems by running InsertCompletionHide when a completion is
accepted _before_ inserting anything else into the buffer. This should
make it much easier to fully implement [LSP text edits]. I doubt
that anyone besides kak-lsp is using the hook parameter today so this
should be a low-risk fix.
[original motivation]: https://github.com/mawww/kakoune/issues/2898
[snippets]: https://github.com/kak-lsp/kak-lsp/pull/616#discussion_r883208858
[LSP text edits]: https://github.com/kak-lsp/kak-lsp/issues/40
This approach is not very elegant as it hooks into the event manager
deep inside the call graph, but solves the exiting issue and is an
okay stop gap solution until a better design comes up.
Fixes#4605
Ubuntu 20.04 ships GCC's libstdc++ 10 from 2020 which implements
std::to_chars() for integers but not for floats. Use the float overload
only if the library advertises support via the feature testing macro.
This can be removed once we require GCC 11 (see
https://www.gnu.org/software/gcc/gcc-11/changes.html).
Closes#4607
gcc 11.2.0 compiles us just fine but clang 13.0.1 fails with this error
clang++ -DKAK_DEBUG -O0 -pedantic -std=c++2a -g -Wall -Wextra -Wno-unused-parameter -Wno-sign-compare -Wno-address -frelaxed-template-template-args -Wno-ambiguous-reversed-operator -MD -MP -MF .ranges.debug.d -c -o .ranges.debug.o ranges.cc
ranges.cc:30:17: error: no viable constructor or deduction guide for deduction of template arguments of 'Array'
check_equal(Array{{""_sv, "abc"_sv, ""_sv, "def"_sv, ""_sv}} | flatten(), "abcdef"_sv);
^
./constexpr_utils.hh:14:8: note: candidate template ignored: couldn't infer template argument 'T'
struct Array
^
./constexpr_utils.hh:14:8: note: candidate function template not viable: requires 0 arguments, but 1 was provided
1 error generated.
The same error can be reproduced with this C++ input
template<typename T, int N>
struct Array
{
T m_data[N];
};
void test() {
(void)Array{{1, 2}};
}
Since "Array" has no constructor, the compiler uses aggregate
initialization. Only recent g++ seems to be smart enough to deduce
template arguments in this case. Help other compilers by adding a
deduction guide. The deduction guide needs to count the array elements
to infer the array size, hence we need to remove braces. Happily,
this is allowed and it's also what std::array does.
Closes#4597
Deleting a buffer resets normal mode on all clients that were
displaing that buffer, but ScopedForceNormalMode that are used
from user mode do not take this possiblity into account on
destruction, which leads to deleting the last normal mode from
the context, ending up with an empty mode stack.
Fixes#3909
This fixes a crash when using kak-lsp with bash-language-server. The
issue is that the second read() in parse_quoted may read past the end of
the string. If this happens and the condition on line 126 is false,
then the loop on line 119 will continue to read past the end of the
buffer since it checks for state.pos != end instead of state.pos < end,
which will likely result in a crash. The fix is to add a check for the
buffer end before the second read. The added test fails without the
change and passes with the change.
Given a completer option with two applicable completions
text|select-cmd1|menu-text1
text|select-cmd2|menu-text2
Kakoune will only show one of them, because they will insert the
same text.
Some language servers send completions like this, for example if
two different importable modules provide the same name. This can be
reproduced using intelephense in this PHP file (cursor is %())
<?php
namespace namespace1;
class sometype {}
?>
<?php
namespace namespace2;
class sometype {}
?>
<?php
namespace test;
some%()
?>
Both completions insert "sometype". The import statement will be
added in an InsertCompletionHide hook by kak-lsp (it uses select-cmd
to determine which completion was selected).
To support this use case, refine the duplicate detection to not filter
out completions with different select-cmd values.
Parsing a (non-valid) font with a comma in the name of the base colour
makes Kakoune crash. It is not a valid face, but Kakoune should just
return an error message instead.
Reproducer:
:set-face global foo ,red@,blue
Note the comma "," after the "@". This is not a valid base name, and it
leads to a crash. Let's see what happens.
At the beginning of parse_face(), we have the following code:
auto bg_it = find(facedesc, ',');
auto underline_it = bg_it == facedesc.end() ? bg_it : std::find(bg_it+1, facedesc.end(), ',');
auto attr_it = find(facedesc, '+');
auto base_it = find(facedesc, '@');
[...]
auto colors_end = std::min(attr_it, base_it);
After this:
- bg_it points to ",red@,blue"
- bg_it != facedesc.end(), so we have underline_it pointing to the first
comma after bg_it. This means that underline_it points to ",blue"
- base_it points to "@,blue"
- attr_it points to the end of facedesc (no "+" marker), so colors_end
points to base_it, "@,blue"
Later in the code, just after parsing the foreground and background
colours, we have:
if (underline_it != facedesc.end())
face.underline = parse_color({underline_it+1, colors_end});
When passing {underline_it+1, colors_end} to parse_color(), we pass in
fact iterators pointing to {",blue", "@,blue"}. Because the second one
starts _before_ the first one in the string, this means that the
resulting string is considered to have a _negative_ length.
parse_color() passes the string to str_to_color(), who fails to turn up
the colour, and attempts to throw:
throw runtime_error(format("unable to parse color: '{}'", color));
The variable "color" still has this negative length, and this goes all
the way down to an assert in src/units.hh where we expect that string to
be >= 0, and we crash on the assertion failure.
For similar reasons, we also get a crash if the comma comes after the
marker for the face attributes:
:set-face global foo ,red+,a
To fix both cases, let's add a check to make sure that the underline_it,
marked with a comma, never gets detected as present and pointing after
colors_end, be it "@" or "+".
As pointed out in [1], when insert mode autocomplete is disabled,
<c-n> could be used to activate insert mode completions temporarily
[2]. This regressed in 6f7c5aed (Do not show custom completions when
autocomplete is off, 2022-01-03). Fix this by enabling completions
on <c-n>/<c-p>. This allows us to remove a special case for explicit
completers.
Alternative behavior (future?): make <c-n> toggle completion like
<c-o>. This can be done today, as suggested by Screwtape on IRC:
map global insert <c-n> %{<c-o><c-n><a-;>:toggle-ctrl-n<ret>}
define-command toggle-ctrl-n %{
hook global InsertCompletionShow .* %{ map window insert <c-n> <c-n> }
hook global InsertCompletionHide .* %{ unmap window insert <c-n> <c-n> }
}
[1] https://github.com/mawww/kakoune/pull/4493#issuecomment-1031189823
[2] <c-n> completion only lives for the lifetime of the completion
menu, whereas <c-o> lasts until you exit insert mode. This means
that autocompletion is much more convenient than <c-n> or <c-x>f,
because those require an explicit completion request for each
path component.
The ThreadedRegexVM implementation does not execute split opcodes as
expected: on split the pending thread is pushed on top of the thread
stack, which means that when multiple splits are executed in a row
(such as with a disjunction with 3 or more branches) the last split
target gets on top of the thread stack and gets executed next (when the
thread from the first split target would be the expected one)
Fixing this in the ThreadedRegexVM would have a performance impact as
we would not be able to use a plain stack for current threads, so the
best solution at the moment is to reverse the order of splits generated
by a disjunction.
Fixes#4519
This will unfortunately break some use case which will require
using wrapper scripts to add the necessary newline. It is however
harder to do the contrary, and it makes a lot of other use case
possible, such as checksuming.
Fixes#3669
After a while it seems clear changing this is much more ergonomic
and restoring it with pure config is impractical as we need to map
all lower case keys.
As reported in [1], completions provided by "set global completers
option=my_completion" activate insert mode autocompletion, even when
the autocomplete option does not have the insert mode flag.
This happens because InsertCompleter::on_option_changed() calls
InsertCompleter::setup_ifn(), which shows the completion pager.
Fix this by computing whether the completion pager is enabled;
otherwise we can return early from setup_ifn().
The completion pager is enabled if the autocompletion bit is set,
or if the user has requested explicit completion.
[1]: https://github.com/kak-lsp/kak-lsp/issues/585
In normal mode, the mode line contains "1 sel" or "n sels (k)" when n > 1,
whereas in insert mode, it contains "n sels (k)" even for n == 1. Change
the contents in insert mode to match normal mode.
trim_indent call was incorrect, trim_indent is intended to work
on multi-line strings and trims trailing whitespace as well (could
benefit from a better name).
Fixes#4378
Merge all lookarounds into the same instruction, merge splits, merge
literal ignore case with literal...
Besides reducing the amount of almost duplicated code, this improves
performance by reducing pressure on the (often failing) branch target
prediction for instruction dispatching by moving branches into the
instruction code themselves where they are more likely to be well
predicted.
Only ui type Terminal is intended to be a user interactive session.
If your ui type is not Terminal, don't worry about making
the tty your stdin if fd 0 is not a tty.
This allows json-rpc commands sent via stdin to be acted up rather
than sent to a fifo (which is the default behavior for kakoune).
Does not change the behavior for Terminal ui sessions
Now that Kakoune opts into extended key reporting, <c-i> is correctly
reported and hence needs to be mapped to forward jump.
We still need to keep <tab> mapped to it for legacy terminals.
Should fix#4333
Closing/reopening the read side seems to sometimes lead to
end-of-file not being received, leaving some extra data unexecuted.
The FDWatcher stays disabled during the executing of the fifo
commands so this should not enable any more commands to be executed
until we get back from the CommandManager and reset the FDWatcher
fd to to fifo read end.
Fixes#4410
ARM uses @ as a comment character, so %progbits must be
used in place of @progbits here. This change fixes the
build on armv7 FreeBSD 13.0.
Fixes mawww/kakoune/issues#4385
See also https://bugs.freebsd.org/259434
Cppcheck produces the following warnings:
```
shared_string.hh:27:49: portability: Shifting signed 32-bit value by 31 bits is implementation-defined behaviour
shared_string.hh:27:49: error: Signed integer overflow for expression '1<<31'.
```
Fixes#4340
Cppcheck produces the following warning:
```
keymap_manager.hh:54:37: performance: Function parameter 'user_mode_name' should be passed by const reference.
```
Fixes#4340
Enable it if supported by default, let the user override it with
the existing terminal_synchronized ui option.
This should finalize work discussed on #4317
They are quite different use cases, and this allow moving InsertMode
to input_handler.hh which is what uses it.
This also cleans up the code as we can get rid of get_insert_pos and
rely more on SelectionList::for_each.
This commit prevents `ga` from returning a “no last buffer” error
when the previously displayed buffer was removed.
Since the jumps list keeps track of the order in which buffers were
displayed already, handling arbitrary `delete-buffer`s as well,
cycle through it to implement `ga` instead of storing a pointer.
Note that this commit doesn't take into account buffer flags that
might exclude some buffers from being cycled over by commands.
Fixes#1840
