Unfortunately Strings that start with a quoted quote (like '''str')
are still incorrectly highlighted, a deeper refactoring of the regions
highlighter will be necessary.
Now that we have a nice standard way to express lists of strings,
registers can be fully exposed. An new $kak_main_reg_... env var
was added to provide the previous behaviour which is relied on by
doc.kak.
Registers are lists of strings, so this make it possible to set
the whole list instead of forcing registers to a single element
when going through the set-register command.
Option lists and maps are specified using separate arguments, avoiding
the need for additional escaping of their separator and reusing the
existing command line spliting logic instead.
As discussed on #2087, this should make it much easier to work with
list options, and make the general option system feel cleaner.
Command line parsing now works as follow:
* Quoted strings ('...', "..." and %~...~ with '~' non nestable)
use 'doubling-up' for escaping their delimiter, if the delimiter
appears twice in a row, it is considered as part of the string and
represent one delimiter character. So 'abc''def' == "abc'def". No
other escaping takes place in those strings.
* Balanced strings (%{...}) do not support any kind of escaping, but
finds the matching closing delimiter by taking nesting into account.
So %{abc{def}} == "abc{def}".
* Non quoted words support escaping of `;` and whitespaces with `\`,
`%`, `'` and '"` can be escaped with `\` at the start of the word,
they do not need escaping (and will not be escaped) else where in
a word where they are treated literally. Any other use of '\' is a
literal '\'. So \%abc%\;\ def == "%abc%; def"
As discussed in #2046 this should make our command line syntax more
robust, provide a simple programmatic way to escape a string content
(s/<delim>/<delim><delim>/g), be well defined instead of ad-hoc
undocumented behaviour, and interact nicely with other common
escaping by avoiding escaping hell (:grep <regex> can in most case
be written with the regex unquoted).
Automatic reparsing of %sh{...}, while convenient in many cases,
can be surprising as well, and can lead to security problems:
'echo %sh{ printf "foo\necho bar" }' runs 'echo foo', then 'echo bar'.
we make this danger explicit, and we fix the 'nop %sh{...}' pattern.
To reparse %sh{...} strings, they can be passed to evaluate-commands,
which has been fixed to work in every cases where %sh{...} reparsing
was used..
This commits changes the way `C` behaves when the next line is empty:
instead of stopping the selection, it will now jump to the next line
that can hold a selection as big as the current one.
The primitive's count parameter holds the maximum amount of selections
that should be added to the current one.
Closes#2061
Seems to work on openbsd 6.3-current but needs more testing. Had to
hardcode the binary path as openbsd considers getting the executable
path at runtime a security flaw.
Window can be resized with an "offset_pos" flag, which means that
the resize took place on the top left corner of the window, leading
to a change in current window position. This is treated as temporary
and the position change is stored in a m_position_offset field.
That allows the ncurses UI to offset the position when it displays
a Search menu, so that the window does not constantly scroll when
the search menu open/closes. The window will only scroll if it needs
to in order to keep the main selectin visible.
We cannot guarantee after an itersel that selections are going to
still be in ascending order, but we were calling a SelectionList
constructor that was assuming this was the case.
Pure clients never quitted when they got SIGHUP after recent changes
to add is_ok to UserInterface. run_client now tracks the UI state as
well and quits if the UI gets killed.
Doing that clears m_buffer_trash, deleting buffers contained there,
but we are not ready to have Buffer destructors running yet as we
did not clear the ClientManager, meaning we might have free windows
pointing to buffers in the buffer trash.
Add a UserInterface::is_ok method and return false on
SIGHUP/stdin closing/socket dropping
This should be cleaner and more robust than the previous SIGHUP
handling code.
Fixes#1594
Instead of using two vectors, we can hold both current and next
threads in a single buffer, with stacks growing on each end.
Benchmarking shows this to be slightly faster, and should use less memory.
Remove ExecState to store threads inside the ThreadedRegexVM so that
memory buffers can be reused between executions. Extract an ExecConfig
struct with all the data thats execution specific to avoid storing
it needlessly inside the ThreadedRegexVM.
If the local client was quitted during its creation (using -e quit
for example), we could have been accessing a null pointer afterwards.
Make the rest of the code aware that local_client might be null.
Shifted function keys are not well standardized around terminals,
Shift F(N) usually returns F(X) + N, with X=12 on xterm, X=10 on
rxvt-unicode... Default to X=12 and make it configuable through
the ncurses_shift_function_key ui_option.
This fixes what #1898 tried to.
Because keyboard layouts vary, the shift-modifier `<s-…>` is only supported
for special keys (like `<up>` and `<home>`) and for ASCII lowercase where
we assume the shift-modifier just produces the matching uppercase character.
Even that's not universally true, since in Turkish `i` and `I` are not an
uppercase/lowercase pair, but Kakoune's default keyboard mappings already
assume en-US mappings for mnemonic purposes.
Mappings of the form `<s-x>` are normalized to `<X>` when `x` is an ASCII
character. `<backtab>` is removed, since we can now say `<s-tab>`.
set-face now takes a scope argument, and faces can be overridden on
a buffer or window basis.
colorscheme apply on global scope, which should be good enough for
now.
Fixes#1411
Schedule it for later, when we get "idle". It currently can lead to
crashed because after the callback, the current mode might be different,
leading to a crash when doing the ModeChange hook call.
Session/Client/User modes names are now requiered to be "identifiers"
they must be in [a-zA-Z0-9_-]. Option names are the same except they
do not allow '-' as they need to be made available through the env vars
and '-' is not supported there.
Fixes#1946
In the end, % is not that painful to work with as its only set seldomly,
and we usually dont need to use expansion at the same time. Moreover, it
just requires a single \ to be escaped.
Fixes#1562
When Kakoune forked the sever to background, the newly converted
to client process (the original client/server process) was not
preserving its previous client name.
Letting any character to be escaped is error prone as it looks like
\l could mean [:lower:] (as it used to with boost) when it only means
literal l.
Fix the haskell.kak file as well.
Fixes#1945
As the computation of word boundary matches is separate from the
actual subsequence matching, we sometimes have candidate that match
as a single word while still having multiple word boundary matches.
For example, with query "expresins", candidate "expressionism's"
will match as single word ("expressins" is a subsequence of
"expressionism"), and will have two word boundaries match (it does
match the last "s", which is considered as a separate word).
This should not be taken into account when compared against
candidate "expresions", which should be considered a better
match.
Fixes#1925
Reference highlighters allow for potential mutual recursion between
highlighters. This is usually fine, but if the recursion happens on
the same buffer range, it means we will recurse infinitely.
Fixes#1920
Pressing Y or N will set the buffer (or window, if it is set at that
scope) autoreload option to the corresponding value, avoiding infinite
loops where a file getting constantly modified prevents the user from
using Kakoune.
To allow more general look arounds out of the actual search range,
pass a second range (the actual subject). This allows us to remove
various flags such as PrevAvailable or NotBeginOfSubject, which are
now easy to check from the subject range.
Fixes#1902
Change the logic of open line commands so that if a selection lies
on the end of line character of the line from which we open a new
line, that selection does not move.
If we have two clients, A and B, with B's cursor on the eol character
of line L, and A hits `o` while on line L, B's cursor should stay
on the same (logical) line. Previous behaviour would make B's cursor
jump on the newly inserted line.
Selection merging is necessary if we want X to work nicely when we are
on EOL (jumping to next line is as nice as it could be, and we are much
more often on EOL nowadays).
Some iterators were refering to to their view with a const ref, this
was making them non-copiable. Change those const ref into const pointers
in order to fix that.
This change displays command-line options in grid format. Each
parameter is indented with two spaces and then padded to maintain
vertical alignment of each description.
I think the visual spacing makes the options much easier to
read. This is particularly important for people new to Kakoune who
use `-help` as a way to become familiar with the program.
Various places in Kakoune code used to modify selections so that
cursors would not lie on an end of line. Remove those to increase
Kakoune's consistency and simplicity.
Now that end of lines are highlighted separately, they should not
be handled specially in most commands.
When on an end of line, certain behaviours can be surprising, for
example delete will join the following line (which makes sense, and
is consistent, but hard to predict if we do not know the cursor is
on and end of line).
As Kakoune is moving more and more towards treating end of lines
as any other character, making it clear when the cursor lies on
them seems like a good way to reduce surprise.
Avoiding end of line is a behaviour we avoid (pun intended) more and more
in Kakoune source code, now that end of lines are regularly selected, it
makes no sense to just to next line when last modification lies on an EOL
(and it probably did not make much sense back when that code was written).
This reverts commit 55621fb4cc.
This should not be necessary as :exec/eval already save those registers
and reset them by default, and it breaks the ability to use those
registers during an eval/exec (as the commands behave differently)
There does not seem to be any reasonable use cases of not collapsing
jumps when the input is not comming from the user. Always collapse
them.
It could make sense to move jump collapsing out of context_wrap as
in general any action not comming directly from the user should
collapse them, at the moment a comment or mapping will not collapse
jumps, which is unfortunate.
Aliases are considered again in command name completion, but only
if they are more than 3 charactes long. This should prevent cluttering
with aliases while still letting long ones being completed.
The keymap mode was passed to on_next_key_with_autoinfo, which
means the mapping was applied directly by the InputHandler.
That led to the first key being interpretted as a mapping, then
all following keys being executed normally, as if typed in the
parent mode of the user mapping.
The debug buffer is a bit special as lots of events might mutate it,
permitting it to be modified leads to some buggy behaviour:
For example, `pipe` uses a ForwardChangeTracker to track buffer
changes, but when applied on a debug buffer with the profile flag
on, each shell execution will trigger an additional modification
of the buffer while applying the changes, leading to an assertion
failing as changes might not be happening in a forward way anymore.
Trying to modify a debug buffer will now raise an error immediatly.
forward (which controls if we are compling for forward or backward
matching) is always statically known, and compilation will first
compile forward, then backward (if needed), so by having separate
compiled function we get rid of runtime branches.
Not only are display columns rarely used to give error positions,
but they make the parsing much slower as for each token we need to
compute the column in the line.
If we reload a buffer, it means its underlying file exists, hence the
New flag does not make sense anymore. It could be that the file appeared
on the filesystem in the meantime.
`:source` command will now generate timings if profile is enabled
in the debug option, to help find which script can be slow to load.
This should help for #1823
Stripping whitespaces there is a failed experiment as it breaks the
ability to use multi-selections consistently: Using '*' followed by some
`N` to add following matches, we end up with mismatched selections
due to whitespace stripping the original selection still contains
whitespaces where all the new ones do not. Once we get to this state,
most selection commands will give different results for the initial
selection and the other ones, breaking predictible multiselection use,
one of the cornerstones of Kakoune editing model.
reset the mouse state so that the terminal can take back control
of the mouse while Kakoune is suspended, and does not emit focus
events anymore.
Fixes#1816
Reseting normal mode will enable normal mode, which will trigger
a check for buffer modification. We do not want that check to
happen as we are trying to close the info box. Doing that mode
reset first will prevent the check from happening (as the info
box is already displayed), and will correctly hide it afterwards.
Fixes#1809
This change is useful when using `set scope completers word=buffer`,
instead of the default word=all.
If candidates all come from the same buffer, then the path/filename
information is the same and therefore unnecessary. This change
prevents the same path from being repeated, and the buffer's
source code is less obscured.
More generally, there could be an option to disable the path
information entirely in all cases, but for now this change seems
a reasonable solution until any such option exists.
Handle next event should never block if we have already accumulated
input that we want to process. As we can accumulate new input in
lots of places (everytime we run a shell process for example, we
might end up reading input keys. That can be triggered during the
mode line generation which takes place during display of the window)
Fixes#1804
Makes it easier for users who want to locate their kakrc file, and
does not require to go through shell expansion to get it as
"${XDG_CONFIG_DIR:-${HOME}/.config}/kak"
Fixes#1740
Always consider that the first selection in the list is the main
one, save selections that way.
This approach was suggested by PR #1786 but the implementation here
is different, and is used more generally whenever we save selections
to strings.
This is also the prefered way to work only on the main selection:
save selections with Z, reduce to main with <space>, restore with z.
Closes#1786Fixes#1750
Do not allocate temporary vectors to store splitted data, use the
'split' range adaptor along with transform(unescape) to provide the
same feature with less allocations.
The words we store in the WordDB are dependent on the extra_word_chars
options, which can be different for different buffers. When completing
words in a buffer based on the WordDB from another buffer, some candidates
might contains characters that are not considered word character for
the target buffer, ignore those words.
No need to have two separate regexes to handle forward and backward
matching, just passing RegexCompileFlags::Backward will add support
for backward matching to the regex. For backward only regex, pass
RegexCompileFlags::NoForward as well to disable generation of
forward matching code.
InsertCompletionSelect will be called whenever the selected insert
completion changes. If the original text is selected back, the hook
parameter will be empty. If another candidate is selected, the hook
parameter will be its text content.
Fixes#1676
Some highlighters, such as wrap or line numbers, are not intended
to be used multiple times on the same display. Add support for unique
ids that are used by highlighters to disable themselves if another
unique highlighter with the same id is supposed to override them.
The usual highlighter "precedence" takes, place, that it, that most
nested highlighter will the the one to run (window in priority to
buffer in priority to global).
* use the list_separator variable instead of hard coding ':'
* fix trailing separator when converting empty prefixed list to string
* correctly escape the prefix in case it contains a separator
Move recording of keys to the input handler itself instead of the
Insert mode so that eventual nested modes (potentially introduced
by <a-;> will get their keys recorded as well).
Fixes#1680
When using an env var that needed the selections in the pipe command line,
say $kak_selection, the selection update code would run, modifying the
selections to adapt to eventual changes. But the rest of the pipe logic
was assuming the selections would not change, leading to bugs.
clamp could change ordering between a coordinate past the end.
Say in a buffer with 1 line of 2 char:
{0, 1} was clamped to {0, 1}
{1, 0} was clamped to {0, 0}
That was reversing their ordering, and might be the root cause
of the bug lurking in undo range computation.
Creating a session will not accept any slashes in the session path,
connecting to an existing session will accept at most one slash to
allow for specifying the session of a different user.
Fixes#1635
AstNodes are now POD, stored in a single vector, accessed through
their index. The children list is implicit, with nodes storing only
the node index at which their child graph ends.
That makes reverse iteration slower, but that is only used for reverse
matching regex, which are uncommon. In the general case compilation
is now faster.
The previous method, which was a bit faster in the general use case,
can hit some cases where we get quadratic behaviour and very slow
matching.
By using an instruction, we can guarantee our complexity of O(N*M)
as we will never have more than N threads (N being the instruction
count) and we run the threads once per codepoint in the subject
string.
That slows down the general case slightly, but ensure we dont have
pathological cases.
This new version is much faster than the previous instruction based
search because it does not use a plain `.*` searcher, but a specific,
smarter instruction specialized for finding the next start if we are
in the correct conditions.
Identify each step with a counter, and check if the instruction
was already processed this step. This makes the matching faster,
by removing the need to maintain a vector of instructions executed
this step.
On big regex, reseting all those flags on all instructions for each
character can become the dominant operation. Track that actual
instructions index processed (the scheduled are already tracked in
the next_threads vector), and only reset these.
