## Summary of the Pull Request
I updated the note prefix for blockquotes with GitHub latest built-in
styles.
## References and Relevant Issues
- https://github.com/orgs/community/discussions/16925
## Detailed Description of the Pull Request / Additional comments
This PR fixed the broken style due to their format change.
## Related former Pull Request
- https://github.com/microsoft/terminal/pull/13615
## Summary
Applications like PowerToys, with their keyboard remapping features
frequently (i.e whenever a remapped shortcut is triggerred) send
`KeyEvent` with out-of-range virtual keycode values (E.g. 0xFF). This is
fixed for WT in #7145, we just needed it in our good ol' `conhost`.
After this PR, Key events with an invalid virtual keycode and
scancode==0 are ignored, and are not added to the `InputBuffer`. Incase,
only virtual keycode is valid but not scancode, we will try to infer the
correct scancode using the virtual keycode mapping.
## References and Relevant Issues
#7145#7064
## Validation Steps Performed
- Triggered a remapped shortcut and verified that `showkey -a` doesn't
output `^@` unexpectedly.
- Key events with an Invalid virtual Keycode and Scancode == 0 are
ignored.
- This PR doesn't include any changes for `WM_[SYS][DEAD]CHAR` messages,
they are left unchanged.
This commit slightly modernizes `CommandHistory` by leaning more heavily
on the STL container functionalities. For one, it uses for-range
iterations to loop through `_commands` instead of using `GetNth`
on every iteration. Another major improvement however is that
the code previously copied entire `CommandHistory` instances out of
the linked list `s_historyLists`, then removed the slot and copied
(not moved!) that instance into the front again. Now it uses the
`splice` function from `std::list` to do it in `O(1)` and virtually
cost-free.
Another major improvement (and the one I'm personally interested in)
is the switch from `SHORT` to `int32_t`. This will greatly simplify
the implementation of the future `COOKED_READ_DATA` class, as the
larger integer type will remove worries about over/underflow.
For instance, we can then just blindly increment/decrement the history
position and then only later clamp it to the expected range.
## Validation Steps Performed
* Existing history tests ✅
* History cycling with F8 ✅
* Navigating history with F7 ✅
This change is a fairly subjective one. It was done because
`IsValidStringBuffer` will very soon be the only function left
in `cmdline.cpp`. Removing it allows removing `cmdline.cpp`.
While the code that replaces it is somewhat tricky, it's also much
more straightforward, as the `IsValidStringBuffer` function didn't
just check if the string buffer is valid - it also retrieved the
pointers to each of the strings contained in the buffer.
## Validation Steps Performed
Exhaustively covered by conhost feature tests ✅
This is a minor cleanup to deduplicate the two ReadConsole methods
and will help with making changes to how `COOKED_READ_DATA` is called.
It additionally changes the initial data payload from a `string_view`
to a `wstring_view` as it is guaranteed to be `wchar_t`.
This matches the current `COOKED_READ_DATA` implementation which
blindly assumes that the initial data consists of `wchar_t`.
Closes#5618
`COOKED_READ_DATA` is a little special and requires cursor navigation
based on the raw (buffered) text contents instead of what's in the
text buffer. This requires the introduction of new helper functions
to implement such cursor navigation. They're made part of `TextBuffer`
as these helpers will get support graphemes in the future.
It also helps keeping it close to `TextBuffer` as the cursor
navigation should optimally behave identical between the two.
Part of #8000.
Adds proper `type` for `SchemePair` definition to avoid warnings about matches of multiple schemas.
Same fix as https://github.com/microsoft/terminal/pull/4045
## Validation Steps Performed
- Pointed $schema to local file instead of https://aka.ms/terminal-profiles-schema
- Confirmed warning goes away when using a string
- Confirmed using the light/dark object format still passes validation
- Confirmed values like `"colorScheme": 3` no longer incorrectly pass validation whereas they would before
When the OS' "text size" setting gets set to 200% and the display
resolution is reduced quite a bit, we get some cropped text in the SUI's
Default Terminal ComboBox. Turns out, we have a height set on the items.
I went ahead and removed that so we don't crop the text. Everything
looks good still!
A similar issue occurs in the Profile > Appearance > Color Scheme
ComboBox. I went ahead and fixed that too by removing the height
restriction.
Other minor changes:
- fixed the comments
- changed "author and version" row to "auto" instead of "*" (star sizing
is great for proportional sizing, so we're not really taking advantage
of it)
Closes#15149
This pull request introduces the arm64 testing agents and a few build
phases to use them.
In addition to running the ARM64 tests in CI, it makes the following
changes:
- The x64 tests now run on equivalent x64 testing agents
- We now run ARM64 builds (and tests!) on all pull requests
- I've deduplicated a lot of the build and test stages
- New queue-time parameters have been added to control various phases,
for quick pipeline testing
- A bunch of conditions have been promoted to compile-time checks to
control the existence of stages and steps more tightly
Adds an `AutomationProperty.Name` to the main grid in the `SettingContainer`. Doing so makes it so that the group of elements is considered a "group \<header\>".
Now, when navigating with a screen reader, when you enter the group of elements, the "group \<header\>" will be presented. Thus, if the user navigates to the "reset" button, it'll be prefaced with a "group \<header\>" announcement first. If the user navigates to it from the other direction (the setting control), this announcement isn't made, but the user already has an understanding of what group of settings they're in, which is standard practice.
Closes#15158
Using our own pools like this gives us a lot of freedom in the tooling
that's installed, the OS versions it targets, and when we take on Visual
Studio updates.
As part of this effort, I've also stood up a "small" agent pool. At the
time of this PR, that pool is using D2ads-v5 SKU VMs (2 vcore 8 GiB)
versus the "large" agent pool's D8as-v5 (8 vcore 32 GiB). Smaller build
tasks have been moved over to the small pool. Compilation's the hard
part, so it gets to stay on the large pool.
`s_TraceApi` was a magic function in Tracing that logged a different
event based on what type it was called with. It was bad for two reasons:
1. I wanted to add a field to each trace indicating the originating
process and thread. This would have required adding a `CONSOLE_API_MSG`
parameter to _every instance_ of `s_TraceApi`, and even then it would
have not been particularly consistent.
2. The design of Tracing, where the TraceLogging macros are hidden
inside opaque functions, subverts the lightweight trace probe detection
present in `TraceLoggingWrite`. Every tracing probe turned into a call
to a cold function which, in 99% of cases, returned immediately.
To that end, I've introduced a new macro _only_ to ApiDispatchers that
emits a named probe with a set of preloaded information. It is a macro
to avoid any unnecessary branching or the emission of any explicit
tracing functions into the final binary.
I have also removed the generic handler for timing any/all API calls, as
we never used them and they were largely redundant with the information
we were capturing from API-specific reports.
I've also removed tracing from all APIs that do not mutate console
state. With the notable exception of ReadConsoleInput, we will see logs
only for things that change mutable console state.
All these things together allows us to construct a process+API-focused
timeline of console events, ala:
```
cmd.exe (20304) CookedRead pwsh 4 07/13/2023 22:02:53.751
cmd.exe (20304) API_GetConsoleMode True
cmd.exe (20304) API_SetConsoleMode False 0x00000003
cmd.exe (20304) API_SetConsoleMode True 0x000001F7
pwsh.exe (4032) ConsoleAttachDetach 07/13/2023 22:03:17.393 True True
pwsh.exe (4032) API_GetConsoleMode False
pwsh.exe (4032) API_GetConsoleMode False
pwsh.exe (4032) API_SetConsoleMode False 0x00000007
```
This pull request also switches the ConsoleAttachDetach and CookedRead
reports to use the PID and FILETIME markings for their pids and
filetimes. This is compatible with the uint32 and uint64 fields that
used to use those names, so anybody who was depending on them will
experience no change in functionality.
I also switched up their order to make them more ergonomic in WPA when
combined with the other API_ tracing (as viewed above.)
This PR adds support for **ITU's T.416 - ODA SGR (38/48)** colour
sequence, which makes use of colon instead of semi-colon as a parameter
separator.
- We use semi-colons as the only parameter separator while sending SGR
color sequences to a ConPTY client. This is to keep backward
compatibility.
- In response to `DECRQSS` query, we have decided to use colons, as the
major usecase for such queries are feature detection (whether client
supports ODA colours), and tracking the original separator may add too
much complexity to the codebase.
## Validation Steps Performed
- Made sure that we are always sending semi-colon separated parameters
regardless of whether the original sequence used colons.
- Made sure that we are always using colons as the parameter separator
in a `DECRQSS` response.
- Added new tests!
Closes#15706
Replace deprecated winrt::apartment_context pattern
I only found 2 instances of this pattern in use and one of them was
actually already replaced but the `co_await winrt::apartment_context`
code was still there.
Tested both window renaming and opening terminal with persisted layouts.
Both still work.
Closes#12982
## Summary of the Pull Request
> ## Abstract
>
> _"Shell integration" refers to a broad category of ways by which a
commandline
> shell can drive richer integration with the terminal. This spec in
particular is
> most concerned with "marks" and other semantic markup of the buffer._
>
> Marks are a new buffer-side feature that allow the commandline
application or
> user to add a bit of metadata to a range of text. This can be used for
marking a
> region of text as a prompt, marking a command as succeeded or failed,
quickly
> marking errors in the output. These marks can then be exposed to the
user as
> pips on the scrollbar, or as icons in the margins. Additionally, the
user can
> quickly scroll between different marks, to allow easy navigation
between
> important information in the buffer.
>
> Marks in the Windows Terminal are a combination of functionality from
a variety
> of different terminal emulators. "Marks" attmepts to unify these
different, but
> related pieces of functionality.
## Detailed Description of the Pull Request / Additional comments
_\*<sup>\*</sup><sub>\*</sub> read the spec
<sub>\*</sub><sup>\*</sup>\*_
In all seriousness, I've already implemented a pile of this. This is
just putting the finishing touches of formalizing it.
## PR Checklist
- [x] This is a spec for #11000 and everything linked to that.
## Summary of the Pull Request
Add support for running profiles in the Add Tab drop down as
administrator without a keyboard.
## References and Relevant Issues
#14517
## Detailed Description of the Pull Request / Additional comments
This pull request adds a FlyoutMenu to each Profile entry in the Add New
tab drop down. When a profile is right clicked or held for 2 seconds in
the case of no mouse input will present a MenuItem to allow the user to
click and run the selected profile as administrator
## Validation Steps Performed
- Responds to pointer input events (mouse, pointer, touchpad)
- Adjusts to theme changes.
- Only shows when a profile is selected. Will not show on settings or
pallete entries
## PR Checklist
- [x] Closes#14517
- [ ] Tests added/passed
- [ ] Documentation updated
- If checked, please file a pull request on [our docs
repo](https://github.com/MicrosoftDocs/terminal) and link it here: #xxx
- [ ] Schema updated (if necessary)
This is my proposed solution to #15384.
Basically, the issue is that we cannot ever close a
`DesktopWindowXamlSource` ("DWXS"). If we do, then any other thread that
tries to access XAML metadata will explode, which happens frequently. A
DWXS is inextricably linked to an HWND. That means we have to not only
reuse DWXS's, but the HWNDs themselves. XAML also isn't agile, so we've
got to keep the `thread` that the DWXS was started on alive as well.
To do this, we're going to introduce the ability to "refrigerate" and
"reheat" window threads.
* A window thread is "**hot**" if it's actively got a window, and is
pumping window messages, and generally, is a normal thing.
* When a window is closed, we need to "**refrigerate**" it's
`WindowThread` and `IslandWindow`. `WindowEmperor` will take care of
tracking the threads that are refrigerated.
* When a new window is requested, the Emperor first try to
"**reheat**"/"**microwave**" a refrigerated thread. When a thread gets
reheated, we'll create a new AppHost (and `TerminalWindow`/`Page`), and
we'll use the _existing_ `IslandWindow` for that instance.
<sub>The metaphor is obviously ridiculous, but _you get it_ so who
cares.</sub>
In this way, we'll keep all the windows we've ever created around in
memory, for later reuse. This means that the leak goes from (~12MB x
number of windows closed) to (~12MB x maximum number of simultaneously
open Terminal windows). It's still not good.
We won't do this on Windows 11, because the bug that is the fundamental
premise of this issue is fixed already in the OS.
I'm not 100% confident in this yet.
* [x] There's still a d3d leak of some sort on exit in debug builds.
(maybe #15306 related)
* havent seen this in a while. Must have been a issue in an earlier
revision.
* [x] I need to validate more on Windows 11
* [x] **BAD**: Closing the last tab on Windows 11 doesn't close the
window
* [x] **BAD**: Closing a window on Windows 11 doesn't close the window -
it just closes the one tab item and keeps on choochin'
* [x] **BAD**: Close last tab, open new one, attempt to close window -
ALL windows go \*poof\*. Cause of course. No break into post-mortem
either.
* [x] more comments
* [ ] maybe a diagram
* [x] Restoring windows is at the wrong place entirely? I once reopened
the Terminal with two persisted windows, and it created one at 0,0
* [x] Remaining code TODO!s: 0 (?)
* [ ] "warm-reloading" `useTabsInTitlebar` (change while terminal is
running after closing a window, open a new one) REALLY doesn't work.
Obviously restores the last kind of window. Yike.
is all about #15384closes#15410 along the way. Might fork that fix off.
Resurrection of #9222.
Spec draft in #9365.
Consensus from community feedback is that the whole of that spec is
_nice to have_, but what REALLY matters is just broadcasting to all the
panes in a tab. So, in the interest of best serving our community, I'm
pushing this out as the initial implementation, before we figure out the
rest of design. Regardless of how we choose to implement the rest of the
features detailed in the spec, the UX for this part of the feature
remains the same.
This PR adds a new action: `toggleBroadcastInput`. Performing this
action starts broadcasting to all panes in this tab. Keystrokes in one
pane will be sent to all panes in the tab.
An icon in the tab is used to indicate when this mode is active.
Furthermore, the borders of all panes will be highlighted with
`SystemAccentColorDark2`/`SystemAccentColorLight2` (depending on the
theme), to indicate they're also active.
* [x] Closes#2634.
- (we should lick a reserved thread for follow-ups)
Co-authored-by: Don-Vito khvitaly@gmail.com
Move scroll marks to `TextBuffer`, so they can be cleared by
EraseInDisplay and EraseScrollback.
Also removes the namespacing on them.
## References and Relevant Issues
* see also #11000 and #15057
* Resize/Reflow _doesn't_ work yet and I'm not attempting this here.
## Validation Steps Performed
* `cls` works
* `Clear-Host` works
* `clear` works
* the "Clear buffer" action works
* They work when there's marks above the current viewport, and clear the
scrollback
* they work if you clear multiple "pages" of output, then scroll back to
where marks previously were
* resizing doesn't totally destroy the marks
Closes#15426
This commit fixes a number of issues around horizontal scrolling.
DxEngine only had one bug, where the clip rect would cause any content
outside of the actual viewport to be invisible. AtlasEngine had more
bugs, mostly around the conversion from textbuffer-relative coordinates
to viewport-relative coordinates, since AtlasEngine stores things like
the cursor position, attributes, etc., relative to the viewport.
It also renames `cellCount` to `viewportCellCount`, because I realized
that it might have to deal with a `textBufferCellCount` or similar in
the future. I hope that the new name is more descriptive of what it
refers to.
Future improvements to AtlasEngine in particular would be to not copy
the entire `Settings` struct every time the horizontal scroll offset
changes, and to trim trailing whitespace before shaping text.
This is in preparation for #1860
## Validation Steps Performed
* Patch `RenderingTests` to run in the main (and not alt) buffer
* Horizontal scrolling of line renditions and attributes works ✅
* Selection retains its position (mostly) ✅
Adds support for colon `:` separated sub parameters in parser.
Technically, after this PR, nothing should change except, now sub
parameters are parsed, stored safely and we don't invalidate the whole
sequence when a `:` is received within a parameter substring.
In this PR:
- If sub parameters are detected with a parameter, but the usage is
unrecognised, we simply *skip* the parameter in `adaptDispatch`.
- A separate store for sub parameters is used to avoid too many changes
to the codebase.
- We currently allow up to `6` sub parameters for each parameter, extra
sub parameters are *ignored*.
- Introduced `VTSubParameters` for easy access to underlying sub
parameters.
> **Info**: We don't use sub parameters for any feature yet, this is
just the core implementation to support newer usecases.
## Validation Steps Performed
- [x] Use of sub parameters must not have any effect on the output.
- [x] Skip parameters with unexpected set of sub parameters.
- [x] Skip sequences with unexpected set of sub parameters.
References #4321
References #7228
References #15599
References https://github.com/xtermjs/xterm.js/pull/2751Closes#4321
This commit inlines `EventsToUnicode` into `WriteConsoleInputAImpl`
because soon we'll not use deques for events anymore and so the old
code won't work. It cleans up the implementation because I intend to
move all this code directly into `InputBuffer` to have a better and
tighter control over how text gets converted. UTF-8 input for instance
requires the storage of up to 3 input events and this code is not fit
to handle that. It's also unmaintainable because our input handling
code shouldn't be spread over a dozen files either. 😄
## Validation Steps Performed
* Unit and feature tests are ✅
Adds a note to the ReadMe's installation instructions which describes
why current versions of Terminal are unavailable via winget.
## PR Checklist
- [x] Closes#15663
- [ ] Tests added/passed
- [ ] Documentation updated
- If checked, please file a pull request on [our docs
repo](https://github.com/MicrosoftDocs/terminal) and link it here: #xxx
- [ ] Schema updated (if necessary)
## Summary of the Pull Request
Fix C2664 errors under latest compiler.
## References and Relevant Issues
#15309
## Detailed Description of the Pull Request / Additional comments
- Latest compilers are more strict
- Internal background of change:
[DevDiv:1810844](https://devdiv.visualstudio.com/DevDiv/_workitems/edit/1810844)
## Validation Steps Performed
- Now successfully builds under VS `17.8.0 Preview 1.0 `
- Still successfully builds under VS `17.6.5`
## PR Checklist
- [x] Closes#15309
- [ ] Tests added/passed
- [ ] Documentation updated
- If checked, please file a pull request on [our docs
repo](https://github.com/MicrosoftDocs/terminal) and link it here: #xxx
- [ ] Schema updated (if necessary)
---------
Co-authored-by: Dan Albrecht <danalb@ntdev.microsoft.com>
I wrote this tool to help me test the buffer reflow code in Windows
Terminal. It needs to stay in sync with the buffer contents in ConPTY
which is somewhat tricky to achieve. This tool will make this easier
because it allows me to quickly, visually compare the contents.
This reverts a number of changes to input handling to how it used to be
in conhost v1. It merges the input event coalescing logic into a single
function and inlines the console suspension event handling, because
soon these functions will receive `std::span` arguments which cannot
be preprocessed anymore, unlike a `std::deque`.
It also adds back support for Ctrl-S being an alias for VK_PAUSE
which was lost in commit fccc7410 in 2018.
Closes#809
## Validation Steps Performed
* Unit and feature tests are ✅
* Ctrl-S pauses output 🎉
The DWMWA for this has been documented for quite a while now!
I've also updated to a version of TerminalThemeHelpers that removes all the Dark Theme exports.
Add prIssueManagement.yml to onboard repo to GitOps.ResourceManagement
as FabricBot replacement
---------
Co-authored-by: microsoft-github-policy-service[bot] <77245923+microsoft-github-policy-service[bot]@users.noreply.github.com>
## Summary of the Pull Request
Adds a dismiss selection option to the "copy" action.
## PR Checklist
- [x] Closes#15371
- [x] Tests added/passed
- [x] Documentation updated
- If checked, please file a pull request on [our docs
repo](https://github.com/MicrosoftDocs/terminal) and link it here:
MicrosoftDocs/terminal#686
- [x] Schema updated (if necessary)
---------
Co-authored-by: Carlos Zamora <carlos.zamora@microsoft.com>
Performance of printing enwik8.txt at the following block sizes:
4KiB (printf): 78MB/s -> 93MB/s
128KiB (cat): 117MB/s -> 156MB/s
The change itself is rather self-explanatory.
A tighter, simpler loop runs faster.
## Validation Steps Performed
Mixed ASCII/Unicode text output looks generally correct. ✅
This is a complete rewrite of the old `WriteCharsLegacy` function
which is used when VT mode is disabled as well as for all interactive
console input handling on Windows. The previous code was almost
horrifying in some aspects as it first wrote the incoming text into a
local buffer, stripping/replacing any control characters. That's not
particular fast and never was. It's unknown why it was like that.
It also measured the width of each glyph to correctly determine the
cursor position and line wrapping. Presumably this used to work quite
well in the original console code, because it would then just copy
that local buffer into the destination text buffer, but with the
introduction of the broken and extremely slow `OutputCellIterator`
abstraction this would end up measuring all text twice and cause
disagreements between `WriteCharsLegacy`'s idea of the cursor position
and `OutputCellIterator`'s cursor position. Emoji input was basically
entirely broken. This PR fixes it by passing any incoming text
straight to the `TextBuffer` as well as by using its cursor positioning
facilities to correctly implement wrapping and backspace handling.
Backspacing over Emojis and an array of other aspects still don't work
correctly thanks to cmdline.cpp, but it works quite a lot better now.
Related to #8000Closes#8839Closes#10808
## Validation Steps Performed
* Printing various Unicode text ✅
* On an fgets() input line
* Typing text works ✅
* Inserting text works anywhere ✅
* Ctrl+X is translated to ^X ✅
* Null is translated to ^@ ✅
This was tested by hardcoding the `OutputMode` to 3 instead of 7.
* Backspace only advances to start of the input ✅
* Backspace deletes the entire preceding tab ✅
* Backspace doesn't delete whitespace preceding a tab ✅
* Backspacing a force-wrapped wide glyph unwraps the line break ✅
* Backspacing ^X deletes both glyphs ✅
* Backspacing a force-wrapped tab deletes trailing whitespace ✅
* When executing
```cpp
fputs("foo: ", stdout);
fgets(buffer, stdin);
```
pressing tab and then backspace does not delete the whitespace
that follows after the "foo:" string (= `sOriginalXPosition`).
`TerminalInput` is configurable, but almost entirely state-less.
As such it isn't helpful that it emits its output via a callback.
It makes tracing the flow of data harder purely from reading the code
and also raises uncertainty about when `TerminalInput` may generate
output. This commit makes it more robust by having `TerminalInput`
simply return its data. Furthermore, it returns that data as a string
instead of converting back and forth between text and `IInputEvent`.
This change will help me make conhost's `InputBuffer` implementation
leaner and help me confidently make more difficult changes to it
with the goal to improve our Unicode support/correctness.
## Validation Steps Performed
* Windows Terminal produces correct results with `showkey -a` ✅
The WPF control has a minor bug where it initializes the renderer
when there isn't even a window yet. When it then calls `SetWindowSize`
it'll pass the result of `GetWindowRect` which is `0,0,0,0`.
This made AtlasEngine unhappy because it restricted the glyph atlas
size to some multiple of the window size. If the window size is `0,0`
then there couldn't be a glyph atlas and so it crashed.
## Validation Steps Performed
* Fixes WPF test control crash on startup ✅
benchcat, "bc" for short, is a tool that I've written over the last
two years to help me benchmark OpenConsole and Windows Terminal.
Initially it only measured the time it took to print a file as fast as
possible, but it's grown to support a number of arguments, including
chunk (`WriteFile` call) sizes, repeat counts and VT mode with italic
and colorized output. In the future I also wish to add a way to
generate the output data on the fly via command line arguments.
One unusual trait of benchcat is that it is compiled entirely without
CRT and vcruntime. I did this so that I could test it on Windows XP.
Also, it's kind of funny seeing how it's only about 11kB.
This commit also fixes a couple `$LASTEXITCODE` cases, because our
spellchecker was bothering me a lot with this PR and so I just fixed it.
The added explicit vectorization allows us to skip plain text faster
and pass it immediately to the deeper `TextBuffer` parts.
Performance of printing enwik8.txt at the following block sizes:
4KiB (printf): 54MB/s -> 58MB/s
128KiB (cat): 103MB/s -> 116MB/s
## Validation Steps Performed
* Works on x64 ✅
* Works on ARM ✅
This fixes a bug reported internally that occurs when resizing the
terminal while also scolling the contents. The easiest way to reproduce
it is to resize the terminal to 0 rows, but it's much more prominent
in a debug build where everything goes out of sync almost immediately.
The underlying issue is that `VtEngine::_wrappedRow` may contain an
offset that is outside of the viewport bounds, because reflowing and
scrolling aren't properly synchronized. The previous `bitmap` code
would then throw an exception for such invalid coordinates and cause
the internal `VtEngine` state to be broken. Once `_wrappedRow` got
to a negative value at least once, it would stay that way unless you're
scrolling up. If the contents are actively scrolling it would quickly
reach a negative value from which it can never recover. At that point
OpenConsole would enter a tight exception-throw-catch-retry loop
and Windows Terminal seemingly cease to show any content.
## Validation Steps Performed
* Resize WT to the minimal window size repeatedly
* Doesn't hang ✅
When we moved the settings UI to lazy initialization in #15628, we broke
PGO. Apparently, we were PGOing the tiny part of Settings that was being
loaded on every launch (e.g. the XAML metadata provider 🤦)
Let's actually PGO launching the settings.
This commit removes some flags that we don't need anymore, and adds all
those `/Zc` (standard conformance) switches that aren't enabled by
default yet. This will help us and the MSVC team detect bugs early.
This removes:
* `/fp:contract`: With the addition of `TIL_FAST_MATH_BEGIN`
all the code that benefits from FMA now uses `/fp:fast`.
* `/Zc:lambda`: Automatically enabled with C++20.
This adds:
* `/Zc:__cplusplus` / `/Zc:__STDC__`: Without these `__cplusplus`
defaults to `199711L` and `__STDC__` remains undefined.
* `/Zc:enumTypes`: The C++ standard specifies that an enum with
unspecifies size has a size that fits its members exactly.
An enum with byte-sized members has a `sizeof` of 1 and not 4.
* `/Zc:templateScope`: Emit errors when shadowing template parameters.
And most importantly:
* `<RemoveUnreferencedCodeData>`, which is `/Zc:inline`
Without this, MSVC treats `inline` functions sort of like external
linkage ones. You can declare an inline function in one file and
then just define it in another. Or use an inline function from
another file. With this flag, the compiler can stop emitting
COMDAT references for these which reduces object file sizes.
Due to an implementation detail in the Xaml compiler--which wants to
ensure that all metadata providers on an App are available
immediately--we were eagerly loading the settings UI DLL and all of its
dependencies, even in sessions where the user was not going to open
Settings.
By turning off eager provider generation and handling it ourselves, we
get to control exactly when the settings UI is loaded.
This required some gentle poking-through of the barrier between App and
Page, but it is almost certainly worth it.
Turning on the Xaml code generation flag to not generate providers
automatically adds an `AddProvider` member to the internal interface for
the autogenerated XamlMetadataProvider. We needed to switch to using the
internal interface rather than the projected type in our custom App base
class to get at it.
Providers that App/Page use must be initialized by the time we start the
WindowsXamlManager, so we load Control and Controls (ha) eagerly and
early.
It looks like it may save 400ms of CPU time (?) on startup.
This commit reduces GdiEngine's average display latency by 8ms,
which caused it to miss a v-blank about half the time at 60Hz.
Closes#15607
## Validation Steps Performed
Input latency with `frarees/typometer` matches conhost from Win10 ✅
UiaRaiseNotificationEvent is not present on Windows Server 2016, even
though it is documented as being present.
This also removes the cost of loading up UIAutomationCore from the
critical path.
This is an improved fix for #13238. Instead of handling focus events in
the `TerminalInput::HandleKey` function and the need to filter them
out depending on where they came from, we simply don't call `HandleKey`
in the first place. This makes the somewhat unreliable `CameFromApi`
function unnecessary and the code a bit more robust.
This change is required because `CameFromApi` is not representable
in a `INPUT_RECORD` and I'm getting rid of `IInputEvent`.
## Validation Steps Performed
* No `[O` when exiting nvim ✅
* Mouse input in nvim works ✅
`(Peek|Read)ConsoleInput(A|W)Impl` make a distinction that doesn't make
a lot of sense in our code base: On the calling side (`ApiDispatchers`)
there's just one function calling all 4 (`ServerGetConsoleInput`) and
on the callee side they all 4 just call `_DoGetConsoleInput` anyways.
## Validation Steps Performed
* It compiles ✅
I've removed these because it made some of my new code pretty
convoluted for now good reason as most of these functions aren't
exception safe to begin with. Basically, their boolean status
is often just a pretense because they can crash or throw anyways.
Furthermore, `WriteCharsLegacy` failed to check the status code
returned by `AdjustCursorPosition` in some of its parts too.
In the future we should instead probably strive to continue
make our legacy code more exception safe.
By rewriting the first major copy loop in `CopyRangeFrom` to use
pointers/iterators instead of indices for iteration, the autovectorizer
kicks in end neatly rewrites it as an unrolled SIMD loop. This improves
performance during traditional window resizes by roughly 2x and will
be quite helpful in the future for our more complex reflow resize.
Unfortunately, MSVC unrolls the loop by 4x which is too much for our
purpose, but there's no option to change that. It's still better than
not having any vectorization however, since it kicks in at 32 columns.
It also renames the function to `CopyTextFrom` be more in line with
the others and to avoid confusion, because it doesn't copy attributes.
## Validation Steps Performed
* Traditional resizing works ✅
LitoMore