This adds support for looking up data under the mouse cursor. Usually it
will bring up a dictionary, but other times it could be a Wikipedia
article, Siri knowledge, etc. Apple doesn't really have a good name for
it, other than "looking up data", "quick look" (a confusingly similar
name with the other Quick Look OS feature), or "show definition". You
can activate this by doing Ctrl-Cmd-D when the mouse is over a cursor.
If you have a trackpad, you can also either activate this using Force
click or three-finger tap (depends on your system preference settings).
Note that for Force click, this could potentially make it impossible to
use the MacVim `<ForceClick>` mapping in Vim, which allows you to map a
force click to a Vim command (#716). This is handled by having a new
setting (under a new "Input" preference pane which will have more
populated later) that allows you to choose whether to use Force click
for data lookup or Vim's `<ForceClick>` mapping. If you have configured
to use three-finger taps though this setting wouldn't do anything, and
`<ForceClick>` is always send to the Vim mapping.
Also, this is lacking a lot of features that a normal macOS application
would get, e.g. looking up selected texts (e.g. if you have "ice cream",
you may want to select the whole thing to look up the phrase, rather
than just "ice" or "cream"), data detector, and much more (e.g. custom
API support). They will be done later as part of #1311.
Technical details below:
The way the OS knows how to look up the data and present it is by
talking to the NSTextInput/NSTextInputClient. Previously MacVim
implemented NSTextInput partially, and didn't implement the critical
firstRectForCharacterRange:actualRange and characterIndexForPoint:
functions. First, in this change we change from NSTextInput to
NSTextInputClient (which is the newer non-deprecated version), and
implement those functions, which allows the OS to query the text
storage.
By default, the OS sends a quickLookWithEvent: call to us whenever the
lookup happens but for some odd reason this isn't automatic for Force
clicks, presumably because some apps want to handle Force clicks
manually (this is why some apps only work for three-finger taps but not
Force clicks for lookups). This isn't documented but I found references
in iTerm/Firefox, and basically we just need to manually handle it and
send off quickLookWithEvent: when handling Force clicks.
For implementing the NSTextInputClient properly, the main issue is
making sure that can work properly with input methods / marked texts,
which is the other primary purpose for this class (other than inputting
keys). For data lookups, I'm representing the grid as a row-major text
(with no newline/space in between) and expose that to the OS. This
already has some issue because it doesn't handle Vim vertical splits
well, as MacVim doesn't really have access to detailed Vim text buffers
easily (unless we do a lot of calls back-and-forth). This means wrapped
texts won't be looked up properly, which I think is ok. Also, the OS
APIs deal with UTF-8 indices, so we can't just convert row/column to raw
indices and have to do a lot of character length calculations
(especially for wide chars like CJK or emojis) to make sure the returned
ranges are consistent and valid. For marked texts though, this presents
a challenge because Vim doesn't really have a strong enough API to
communicate back-and-forth about the marked positions and whatnot (it
only let the GUI know where the current cursor is), and it's hard to
implement APIs like `markedRange` properly because some marked texts
could be hidden or wrapped (if you implement some of these functions
improperly Apple's input methods could start misbehaving especially when
you use arrow keys to navigate). In the end I kept the original
implementation for treating the marked texts as a range starting from 0,
*only* when we have marked text. Kind of a hack but this makes sure we
work both in marked text mode (i.e. when inputting texts) and when doing
lookups. For simplicity I made it so that you can't do data lookups when
in marked text mode now.
Input method:
This change also fixes a quirk in input method as a driveby change.
Previously the logic for calculating the rect for where the candidate
list was quite broken, but now it's calculated correctly using the
desired range and the current cursor position. This matters when say
using Japanese IM and using the left/right arrow to jump to different
sections of the text. If the desired range is in a wrapped line, the new
logic would attempt to pin it to the left-most column of where the
cursor is in the range.
Data detection:
Note that the default implementation is quite bare, and lacks a lot of
smart data detection. For example, if you put your mouse over a URL, it
won't properly select the whole URL, and addresses and dates for example
also won't get grouped together properly. This is because these require
additional implementation (e.g. using NSDataDetector) instead of coming
"for free", and will be handled later. In fact, Apple's WebKit and
NSTextView cheats by calling an internal API framework called "Reveal"
(which you can find out by intercepting NSTextView's calls and/or
looking at WebKit's source code) which is much more powerful and
supports looking up package tracking, airline info, and more, but it's
not available to third-party software (that's why Safari's lookup is so
much better than Chrome/Firefox's).
This isn't tested right now. Future task needs to add XCTest support to
properly test this as there are a lot of edge cases involved here.
Fix#1191
Part of Epic #1311, which contains other items to be implemented.
Since Vim upstream added support for `:confirm q` for terminals, we can
now get our File.Close menu to work properly for them as well. This also
means Cmd-W (which is bound to that menu item) works too.
Also just add validateMenuItem: for MMAppController even if we are just
returning YES. This makes it semantically clearer and is easier to
debug.
Problem: :confirm does not work properly for a terminal buffer.
Solution: Handle :confirm for a terminal buffer differently. (Yee Cheng
Chin, closes#11312)
Problem: There is no real need for a "big" build.
Solution: Move common features to "normal" build, less often used features
to the "huge" build. (Martin Tournoij, closes#11283)
Problem: Spacing-combining characters handled as composing, causing text to
take more space than expected.
Solution: Handle characters marked with "Mc" not as composing.
(closes#11282
Problem: Too many #ifdefs.
Solution: Graduate the +cmdwin feature. Now the tiny and small builds are
equal, drop the small build. (Martin Tournoij, closes#11268)
Problem: Cannot specify another character to use instead of '@' at the end
of the window.
Solution: Add "lastline" to 'fillchars'. (Martin Tournoij, closes#11264,
closes#10963)
Problem: The 'splitscroll' option is not a good name.
Solution: Rename 'splitscroll' to 'splitkeep' and make it a string option,
also supporting "topline". (Luuk van Baal, closes#11258)
Problem: Evaluating "expr" options has more overhead than needed.
Solution: Use call_simple_func() for 'foldtext', 'includeexpr', 'printexpr',
"expr" of 'spellsuggest', 'diffexpr', 'patchexpr', 'balloonexpr',
'formatexpr', 'indentexpr' and 'charconvert'.
Problem: matchaddpos() can get slow when adding many matches.
Solution: Update the next available match ID when manually picking an ID and
remove check if the available ID can be used. (idea by Rick Howe)
