ConPTY: Fix missing flush on console mode changes #15991
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
microsoft-github-policy-service
bot
added
Issue-Bug
zadjii-msft
approved these changes
Sep 21, 2023
DHowett
approved these changes
Sep 21, 2023
This was referenced Oct 1, 2023
microsoft-github-policy-service bot
pushed a commit
that referenced
this pull request
Nov 21, 2023
This changeset fixes an issue caused by #15991 where "chunked" escape sequences would get corrupted. The fix is to simply not flush eagerly anymore. I tried my best to keep the input lag reduction from #15991, but unfortunately this isn't possible for console APIs. Closes #16079 ## Validation Steps Performed * `type ascii.com` produces soft font ASCII characters ✅
DHowett
pushed a commit
that referenced
this pull request
Dec 6, 2023
This changeset fixes an issue caused by #15991 where "chunked" escape sequences would get corrupted. The fix is to simply not flush eagerly anymore. I tried my best to keep the input lag reduction from #15991, but unfortunately this isn't possible for console APIs. Closes #16079 ## Validation Steps Performed * `type ascii.com` produces soft font ASCII characters ✅ (cherry picked from commit bdf2f6f) Service-Card-Id: 91283205 Service-Version: 1.19
github-merge-queue bot
pushed a commit
that referenced
this pull request
Mar 28, 2024
I've found that #16079 was never properly addressed (it still randomly occurred after even after PR #16349), which later led to the issues described in #16769 (nushell flickering due to too many flushes). The crux of the fix is that this brings back the `_noFlushOnEnd` flag that was removed in PR #15991. This is then combined with a change to the cork API: An `uncork` on `VtEngine` now only flushes if `_Flush` got called while it was corked in the first place. `_noFlushOnEnd` prevents us from flushing in between two "unknown" VT sequences (like soft fonts or FTCS) which prevents them from being corrupted. The corking prevents the remaining cases of flushing too often. Long-term, a proper fix would be to pass through VT unmodified. Closes #16769
DHowett
pushed a commit
that referenced
this pull request
Apr 16, 2024
I've found that #16079 was never properly addressed (it still randomly occurred after even after PR #16349), which later led to the issues described in #16769 (nushell flickering due to too many flushes). The crux of the fix is that this brings back the `_noFlushOnEnd` flag that was removed in PR #15991. This is then combined with a change to the cork API: An `uncork` on `VtEngine` now only flushes if `_Flush` got called while it was corked in the first place. `_noFlushOnEnd` prevents us from flushing in between two "unknown" VT sequences (like soft fonts or FTCS) which prevents them from being corrupted. The corking prevents the remaining cases of flushing too often. Long-term, a proper fix would be to pass through VT unmodified. Closes #16769 (cherry picked from commit 1ede023) Service-Card-Id: 91965217 Service-Version: 1.20
DHowett
pushed a commit
that referenced
this pull request
Apr 16, 2024
I've found that #16079 was never properly addressed (it still randomly occurred after even after PR #16349), which later led to the issues described in #16769 (nushell flickering due to too many flushes). The crux of the fix is that this brings back the `_noFlushOnEnd` flag that was removed in PR #15991. This is then combined with a change to the cork API: An `uncork` on `VtEngine` now only flushes if `_Flush` got called while it was corked in the first place. `_noFlushOnEnd` prevents us from flushing in between two "unknown" VT sequences (like soft fonts or FTCS) which prevents them from being corrupted. The corking prevents the remaining cases of flushing too often. Long-term, a proper fix would be to pass through VT unmodified. Closes #16769 (cherry picked from commit 1ede023) Service-Card-Id: 91965216 Service-Version: 1.19
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Previously, all unknown escape sequences would lead to an immediate call
to
VtEngine::_Flush(). This lead to problems with nushell which usesFTCS marks that were unknown to us. Combined with the linewise redrawing
that nushell does, Terminal would get the prompt in two separate frames,
causing a slight flickering.
#14677 fixed this by suppressing the
_Flush()call when unknownsequences are encountered. Unfortunately, this triggered a bug due
to our somewhat "inconsistent" architecture in conhost:
XtermEngine::WriteTerminalWisn't just used to flush unknown sequencesbut also used directly by
InputBuffer::PassThroughWin32MouseRequestto write its mouse sequence directly to the ConPTY host.
VtEnginealready contains a number of specialized member functionslike
RequestWin32Input()to ensure that_Flush()is calledimmediately and another member could've been added to solve this issue.
This commit now adds
RequestMouseModein the same vein.But I believe we can make the system more robust in general by using
eager flushing by default (= safe), similar to how a
write()on aTCP socket flushes by default, and instead only selectively pause and
unpause flushing with a system similar to
TCP_CORK.This seems to work fairly well, as it solves:
In particular the last point is noteworthy, as this commit removes
the last performance bottleneck in ConPTY that isn't
VtEngine.Around ~95% of all CPU and wall time is spent in there now and any
improvements to
VtEngineshould yield immediately results.Closes #15711
Validation Steps Performed