Fix cursor positioning when jumping more than one cursor location at a
time, for example when pressing "ctrl+a" or "ctrl+e" to go to beginning
and end of the line
Add two helps methods to textinput:
1. inserSliceAtCursor allows inserting a slice of bytes at the cursor
position. This allows a program to insert a string at the cursor
position and allow the internal state of the textinput to properly
track where the cursor should be
2. sliceToCursor: allow users to obtain an (allocated) slice of the
content from the beginning of the input to the cursor position
Request a device status report to trigger a read. Wait for the thread to
join in the main loop to ensure that the read thread has fully closed.
This should prevent the need for polling and a quit_fd, and a separate
mechanism on macos.
The render method always hides the cursor, and only shows it if it needs
to be shown. This has the effect that if a caller never shows the
cursor, exits the application, and the shell doesn't re-show the cursor
then we have no visible cursor.
Always show the cursor on deinit to prevent such a curse.
Fixes: #10
I don't know the extent to which you care about accuracy for rendering
statistics. I found when working on seamstress that the `timestamp`
calls in `std.time` were so unreliable as to be unusable, sometimes
reporting negative time deltas within a single
function. `std.time.Timer`, on the other hand, has been rock-solid in
my experience.
Don't print 0 width graphemes or words. Usually these will be
overwritten since we advance by 0 columns, however if one is at the end
of text it can mess up rendering.
Signed-off-by: Tim Culverhouse <tim@timculverhouse.com>
Implement a way for an application to poll the event loop in a blocking
way without popping an event. Implement a non-blocking pop. These
together enable an application to poll the event loop and then drain it.
This is useful when lots of events are delivered in a short amount of
time so an application can batch process the events and then render.
Signed-off-by: Tim Culverhouse <tim@timculverhouse.com>
the man page for macOS `poll(2)` says, under "bugs", that polling
devices does not work, and indeed, although the current implementation
based on poll(2) does produce output, it does not appear to actually
be polling for this reason; instead the wait is blocking on the `read`
call, causing the macOS user to need to input another character to
exit.
this change introduces a wrapper over macOS's implementation of
`select`, modeled after `std.io.poll` as `select.zig`. it is a compile
error to use `select.zig` when `builtin.os.tag.isDarwin()` is false. a
lightly altered version of `Tty.zig` accompanies this change. it's
certainly possible to incorporate the two into one file; i didn't just
to leave the other one in its original state.
with this change, writing to the pipe correctly exits.
additionally, on macOS, `Thread.setName` must be called from the
thread whose name you wish to set, so I have just commented out that
line.
Explicitly reposition the cursor when moving to a new row to enable
terminals that support text selection to _not_ select the entire screen,
since the entire screen is wrapped.
The render loop could get stuck in an infinite loop when it encounters a
zero width cell. Always set the width to at least 1, and assert this.
Signed-off-by: Tim Culverhouse <tim@timculverhouse.com>
