Since now bufpane handles mouse move and release events generically and
separately from mouse press events, that creates a mess when we dispatch
a mouse press event to an inactive pane without making it active.
For example:
1. Click the right button on an inactive pane. It remains inactive.
2. Then click the left button on it. It becomes active, and an
unexpected text selection appears.
The reason is that the release event for the first click was dispatched
to a wrong pane - the (then) active one, so the (then) inactive pane
didn't get the release event and treats the second click not as a mouse
press but as a mouse move.
The simplest way to fix it is to avoid this scenario entirely, i.e.
always activate the pane when clicking any mouse button on it, not just
the left button.
For mouse wheel motion events we keep the existing behavior: the pane
gets the event but doesn't become active. Mouse wheel motion events are
not affected by the described issue, as they have no paired "release"
events.
When we temporarily disable the screen (e.g. during TermMessage or
RunInteractiveShell), if the mouse is pressed when the screen is still
active and then released when the screen is already stopped, we aren't
able to catch this mouse release event, so we erroneously think that the
mouse is still pressed after the screen is restarted. This results in
wrong behavior due to a mouse press event treated as a mouse move event,
e.g. upon the left button click we see an unexpected text selection.
So need to reset the mouse release state to "released" after restarting
the screen, assuming it is always released when the screen is restarted.
If we press mouse, drag and then release, the release event is
generated twice, since both mouse press and mouse drag events have been
saved in mousePressed map. To fix that, ensure that we only store mouse
press events in it.
Introduce separate mouse release and mouse drag (move while pressed)
events: MouseLeftRelease, MouseLeftDrag, MouseRightRelease etc,
to allow binding them to actions independently from mouse press events
(MouseLeft, MouseRight etc).
This change:
- Makes it possible to handle mouse release and drag for arbitrary mouse
events and actions (including Lua actions), not just for MouseLeft as
in the current code.
- Fixes issue #2599 with PastePrimary and MouseMultiCursor actions:
selection is pasted not only when pressing MouseMiddle but also when
moving mouse with MouseMiddle pressed; similarly, a new multicursor is
added not only when pressing Ctrl-MouseLeft but also when moving mouse
with Ctrl-MouseLeft pressed.
My initial approach was not to introduce new events for mouse release
and mouse drag but to pass "mouse released" info to action functions
in addition to *tcell.EventMouse to let the action functions do the
necessary checks (similarly to what MousePress is already doing). But
then I realized it was a bad idea, since we still want to be able also
to bind mouse events to regular key actions (such as PastePrimary)
which don't care about mouse event info.
Adds config option `multimode`, which takes values `tab`, `vsplit`,
or `hsplit` (corresponding to the file-opening commands). I mean to
use it with a command line like
micro -multimode vsplit foo.h foo.c
to open files in a side-by-side split, but if one really wanted to
one could set it in the config file to change the default behavior of
opening multiple files in tabs.
* Improve buffer view relocation after jumping to a far-away location
When the cursor is moved to a location which is far away from the
current location (e.g. after a search or a goto line), the buffer view
is always relocated in such a way that the cursor is at the bottom or
at the top (minus scrollmargin), i.e. as if we just scrolled to this
location. It's not like in other editors, and IMHO it's annoying. When
we jump to a new location far away, we usually want to see more of its
context, so the cursor should be placed closer to the center of the
view, not near its edges.
This change implements the behavior similar to other editors:
- If the distance between the new and the old location is less than one
frame (i.e. the view either doesn't change or just slightly "shifts")
then the current behavior remains unchanged.
- Otherwise the current line is placed at 25% of the window height.
* Postpone calling onBufPaneOpen until the initial resize
It is currently not possible to find out the geometry of a newly created
bufpane in onBufPaneOpen lua callback: bp:GetView() returns {0,0,0,0}
instead of the actual window. The reason is that the bufpane view is not
properly initialized yet when the bufpane is created and the callback is
triggered. It is initialized a bit later, at the initial resize.
So postpone calling onBufPaneOpen until after the initial resize.
* Improve buffer view relocation when opening a file at a far-away location
When a file is opened with the initial cursor location at a given line
which is far away from the beginning of the file, the buffer view is
relocated so that the cursor is at the bottom (minus scrollmargin)
as if we just scrolled to this line, which is annoying since we'd rather
like to see more of the context of this initial location.
So implement the behavior similar to the earlier commit (which addresses
a similar issue about jumping far away after a search or goto):
- If the initial cursor location is less than one frame away from the
beginning of the buffer, keep the existing behavior i.e. just display
the beginning of the buffer.
- Otherwise place the cursor location at 25% of the window height.
It should not return false immediately when no matching brace is found. This makes the jump fails in certain case: `[ )I]` =/=> `[I )]`.
When there is no brace near the cursor, the last statement is also executed. This may cause problems when chaining commands.
* Support for highlighting all search matches (hlsearch)
hlsearch is implemented efficiently using the buffer's line array,
somewhat similarly to the syntax highlighting.
Unlike the syntax highlighter which highlights the entire file,
hlsearch searches for matches for the displayed lines only.
Matches are searched when the given line is displayed first time
or after it was modified. Otherwise the previously found matches
are used.
* Add UnhighlightSearch action
and add it to the list of actions triggered by Esc key by default.
* Add comment explaining the purpose of search map
* Add hlsearch colors to colorschemes
Mostly just copied from the corresponding original (mostly vim) colorschemes.
* Highlight matches during/after replace as well
As a side effect it also changes the last search value, i.e. affects FindNext
and FindPrevious, but it's probably fine. In vim it works the same way.
* Improve hlsearch option description
The new feature of prefilling the search bar with the selected text (added
in 3d0b5db) has an annoying side effect: if we do have some text selected
but we want to search for some other pattern, not the selected text,
then we have to manually delete the prefilled text before we can start
entering our wanted search pattern.
A simple solution is to select the prefilled text, so that we can replace it
with our pattern right away just by typing, without any additional keystrokes.
BufView returns not only the buffer's width and height but also its
x,y position. It may be useful e.g. for checking if a mouse click was
on the actual buffer or ourside it, e.g. on the gutter.
Modified behavior of CursorUp, CursorDown, CursorPageUp etc:
if softwrap is enabled, cursor moves by visual lines, not logical lines.
TODO: implement it also for Home and End keys: move cursor to the
visual start or end of a line. I haven't implemented it for now, because
I'm not sure what should be the behavior of StartOfTextToggle then
(considering that Home key is bound to StartOfTextToggle by default).
Fixes#1598
Fixes issue with the usage of a slightly incorrect buffer height value
(v.Height should be v.Height-1 if statusline is displayed).
Also, to avoid too many duplications, the code reorganized a little:
buffer display params (width, height, gutter offset and others) are
calculated in a single place.
Softwrap implementation enhanced to fix various issues with scrolling,
centering, relocating etc.
The main idea is simple: work not with simple line numbers but
with (Line, Row) pairs, where Line is a line number in the buffer
and Row is a visual line (a row) number within this line.
The logic remains mostly the same, but simple arithmetic operations
on line numbers are replaced with corresponding operations on
(Line, Row) pairs.
Fixes#632, #1657
Fixed regression: since merging keybindings branch, TryBindKey and
UnbindKey and accordingly "bind" and "unbind" commands don't work
(fail to unmarshal bindings.json).
This is just a quick fixup to make TryBindKey and UnbindKey work again.
They still work with "buffer" bindings only.
In JumpToMatchingBrace, the loop should stop immediately after finding the matching bracket.
It causes multiple jumps in certain situations:
`(I [ ]{ }) => ( I[ ]{ })`
This commit exposes the separate infopane bindings to configuration
from the user. This also adds support for separate bindings in the
terminal emulator view. Default bindings are provided, but can also
be rebound in bindings.json.
This commit separates actions in the command bar from actions in
a normal buffer, and implements what is needed to allow rebinding,
although an interface for command bar keybindings is not yet exposed
to the user.
This commit adds support for binding key sequences such as
"<Ctrl-x><Ctrl-c>". This commit does not solve the problem
of global bindings yet, and therefore the command bar doesn't
work properly in this commit.
