I finally got access to a computer using the Microsoft's Windows OS.
That's the opportunity to find and mitigate all the problems
encountered. This patch:
1. Introduce an option and a C++ definition to enable fallback for
Microsoft's terminal emulators. This allows me to see/test the
Microsoft output from Linux. This also allows Windows users to remove
the fallback and target non Microsoft terminals on Windows if needed.
2. Microsoft's terminal suffer from a race condition bug when reporting
the cursor position:
https://github.com/microsoft/terminal/pull/7583.
The mitigation is not to ask for the cursor position in fullscreen
mode where it isn't really needed and request it less often.
This fixes: https://github.com/ArthurSonzogni/FTXUI/issues/136
3. Microsoft's terminal do not handle properly hidding the cursor. Instead
the character under the cursor is hidden, which is a big problem. As
a result, we don't enable setting the cursor to the best position for
[input method editors](https://en.wikipedia.org/wiki/Input_method),
It will be displayed at the bottom right corner.
See:
- https://github.com/microsoft/terminal/issues/1203
- https://github.com/microsoft/terminal/issues/3093
4. Microsoft's terminals do not provide a way to query if they support
colors. As a fallback, assume true colors is supported.
See issue:
- https://github.com/microsoft/terminal/issues/1040
This mitigates:
- https://github.com/ArthurSonzogni/FTXUI/issues/135
5. The "cmd" on Windows do not properly report its dimension. Powershell
works correctly. As a fallback, use a 80x80 size instead of 0x0.
6. There are several dom elements and component displayed incorrectly,
because the font used is missing several unicode glyph. Use
alternatives or less detailled one as a fallback.
Allow Container::Vertical and Container::Horizontal to have an
external selector, similar to Container::Tab.
This is useful for implementing a menu of menu.
Co-authored-by: ArthurSonzogni <sonzogniarthur@gmail.com>
Modify the ftxui::Pixel. Instead of storing a wchar, store a
std::wstring. Now a single pixel can store multiple codepoints.
If a codepoint is of size <=0, it will be appended to the previous
pixel.
Only ftxui::text() is supported. ftxui::vtext support still needs to be
added.
This causes the following CPU and memory regression:
- Memory: Pixel size increases by 200% (16 byte => 48byte).
- CPU: Draw/Second decrease by 62.5% (16k draw/s => 6k draw/s on 80x80)
Both regressions are acceptable. There are still two orders of magnitude
(100x) before the levels where performance/memory concerns begins.
This fixes: https://github.com/ArthurSonzogni/FTXUI/issues/109
Current state: the fuzzer find interesting input immediately...
```
terminate called after throwing an instance of 'std::range_error'
what(): wstring_convert::from_bytes
```
See: https://github.com/ArthurSonzogni/FTXUI/issues/118
There was some undefined behavior to be fixed in the terminal input
parser.
The behavior of flush seems to have change. The fix was to invert '\0'
and std::flush.
wingdi.h (included via Windows.h) defines an RGB macro that breaks
things. If a user really wants that macro in the same file as FTXUI they
can move the Windows.h include to after the inclusion of FTXUI's
headers.
The ESC key generates sequences that are prefix of others. For instance:
- ESC => [27]
- F1 => [27, 79, 8]
As a result, we can't generate the ESC event when receiving [27],
because it might be the start of the [27, 79, 8] sequence (or not).
Application usually applies a timeout to help detecting the ESC key.
This patch introduce a timeout. It is set to 50ms.
Bug: https://github.com/ArthurSonzogni/FTXUI/issues/55
