This commit reduces the code surface that interacts with raw JSON data,
reducing code complexity and improving maintainability.
Files that needed to be changed drastically were additionally
cleaned up to remove any code cruft that has accrued over time.
In order to facility this the following changes were made:
* Move JSON handling from `CascadiaSettings` into `SettingsLoader`
This allows us to use STL containers for data model instances.
For instance profiles are now added to a hashmap for O(1) lookup.
* JSON parsing within `SettingsLoader` doesn't differentiate between user,
inbox and fragment JSON data, reducing code complexity and size.
It also centralizes common concerns, like profile deduplication and
ensuring that all profiles are assigned a GUID.
* Direct JSON modification, like the insertion of dynamic profiles into
settings.json were removed. This vastly reduces code complexity,
but unfortunately removes support for comments in JSON on first start.
* `ColorScheme`s cannot be layered. As such its `LayerJson` API was replaced
with `FromJson`, allowing us to remove JSON-based color scheme validation.
* `Profile`s used to test their wish to layer using `ShouldBeLayered`, which
was replaced with a GUID-based hashmap lookup on previously parsed profiles.
Further changes were made as improvements upon the previous changes:
* Compact the JSON files embedded binary, saving 28kB
* Prevent double-initialization of the color table in `ColorScheme`
* Making `til::color` getters `constexpr`, allow better optimizations
The result is a reduction of:
* 48kB binary size for the Settings.Model.dll
* 5-10% startup duration
* 26% code for the `CascadiaSettings` class
* 1% overall code in this project
Furthermore this results in the following breaking changes:
* The long deprecated "globals" settings object will not be detected and no
warning will be created during load.
* The initial creation of a new settings.json will not produce helpful comments.
Both cases are caused by the removal of manual JSON handling and the
move to representing the settings file with model objects instead
## PR Checklist
* [x] Closes#5276
* [x] Closes#7421
* [x] I work here
* [x] Tests added/passed
## Validation Steps Performed
* Out-of-box-experience is identical to before ✔️
(Except for the settings.json file lacking comments.)
* Existing user settings load correctly ✔️
* New WSL instances are added to user settings ✔️
* New fragments are added to user settings ✔️
* All profiles are assigned GUIDs ✔️
This pull request introduces Microsoft.Terminal.Core.Color as an
alternative to both Windows.UI.Color and uint32_t/COLORREF in the
TerminalCore, ...Control, ...SettingsModel and ...SettingsEditor layers.
M.T.C.Color is trivially convertible to/from til::color and therefore
to/from COLORREF, W.U.Color, and any other color representation we might
need².
I've replaced almost every use of W.U.Color and uint32_t-as-color in the
above layers, with minor exception¹.
The need for this work is twofold.
First: We cannot bear a dependency from TerminalCore (which should,
on paper, be Windows 7 compatible) on Windows.UI or any other WinRT
namespace.
This work removes one big dependency on Windows.UI, but it does not go
all the way.
Second: TerminalCore chose to communicate mostly in packed uint32s
(COLORREF), which was inherently lossy and dangerous.
¹ The UI layers (TerminalControl, TerminalApp) still use
Windows.UI.Color as they are intimately connected to the UWP XAML UI.
² In the future, we might even be able to *use* the alpha channel...
## PR Checklist
* [x] I ran into the need for this when I introduced cursor inversion
* [X] Fixes a longstanding itch
## Validation Steps Performed
Built and ran all tests for the impacted layers, even the local ones!
This is not going to be our plan of record for Universal going forward.
This updates the Universal configuration to 1) match non-universal and 2) switch to local applications
Clang (10) has no trouble optimizing the COLORREF conversion operator to
a simple 32-bit load with mask (!) even though it's a series of bit
shifts across multiple struct members.
MSVC (19.24) doesn't make the same optimization decision, and it emits
three 8-bit loads and some shifting.
In any case, the optimization only applies at -O2 (clang) and above.
In this commit, we leverage the spec-legality of using unions for type
conversions and the overlap of four uint8_ts and a uint32_t to make the
conversion very obvious to both compilers.
x86_64 msvc | O0 | O1 | O2
------------|----|----|--------------------
shifts | 12 | 11 | 11 (fully inlined)
union | 5 | 1 | 1 (fully inlined)
x86_64 clang | O0 | O1 | O2 + O3
-------------|----|----|--------------------
shifts | 14 | 5 | 1 (fully inlined)
union | 9 | 3 | 1 (fully inlined)
j4james brought up some concerns about til::color's minor wastefulness
in https://github.com/microsoft/terminal/pull/7578#discussion_r487355989.
This is a clear, simple transformation that saves us a few instructions
in a relatively common case, so I'm accepting a micro-optimization even
though we don't have data showing this to be a hot spot.
## Summary of the Pull Request
This adds a first iteration on the command palette. Notable missing features are:
* Commandline mode: This will be a follow-up PR, following the merge of #6537
* nested and iterable commands: These will additionally be a follow-up PR.
This is also additionally based off the addenda in #6532.
This does not bind a key for the palette by default. That will be done when the above follow-ups are completed.
## References
* #2046 - The original command palette thread
* #5400 - This is the megathread for all command palette issues, which is tracking a bunch of additional follow up work
* #5674 and #6532 - specs
* #6537 - related
## PR Checklist
* [x] Closes#2046
- incidentally also closes#6645
* [x] I work here
* [x] Tests added/passed
* [ ] Requires documentation to be updated - delaying this until it's more polished.
## Detailed Description of the Pull Request / Additional comments
* There's a lot of code for autogenerating command names. That's all in `ActionArgs.cpp`, because each case is so _not_ boilerplate, unlike the rest of the code in `ActionArgs.h`.
## Validation Steps Performed
* I've been playing with this for months.
* Tests
* Selfhost with the team
This pull request moves swaths of Cascadia to use `til::color` for color
interop. There are still some places where we use `COLORREF`, such as in
the ABI boundaries between WinRT components.
I've also added two more til::color helpers - `with_alpha`, which takes
an existing color and sets its alpha component, and a
`Windows::UI::Color` convertor pair.
Future direction might include a `TerminalSettings::Color` type at the
idl boundary so we can finally stop using UInt32s (!) for color.
## Validation Steps Performed
Tested certain fragile areas:
* [x] setting the background with OSC 11
* [x] setting the background when acrylic is in use (which requires
low-alpha)
til::color will help us move away from COLORREF internally. It supports
conversion to/from COLORREF, and from all types of structs containing
members named R, G, B and A (or r, g, b, and a).
## Validation Steps Performed
Tests; run through profile/colorScheme deserialization with `til::color`
instead of `uint32_t` or `COLORREF`.
