perf: rotate line slices in Insert/DeleteLineArea full-width fast path#119
Open
LXXero wants to merge 1 commit into
Open
perf: rotate line slices in Insert/DeleteLineArea full-width fast path#119LXXero wants to merge 1 commit into
LXXero wants to merge 1 commit into
Conversation
Scrolling (DeleteLineArea at the top of the scroll region) shifted cells one by one — O(rows*cols) Cell struct copies per scrolled line. During bulk output every newline pays this, and it profiled at ~80% of total process CPU in a real terminal emulator (xerotty) flooding 2M lines. Full-width areas (the common case — whole-screen scroll regions) now rotate the line slice headers instead: O(rows) slice-header moves plus clearing the n recycled lines, reusing their storage. Cleared lines are filled by direct assignment — Line.Set's wide-cell repair has nothing to repair when every cell is replaced. Partial-width areas keep the shift but use copy() per row instead of per-cell assignment. Measured end-to-end (seq 1 2000000 into an 80x24 emulator): 31s -> 13.5s wall just from this change; combined with a scrollback ring buffer in x/vt it reaches ~6-8s.
LXXero
added a commit
to LXXero/xerotty
that referenced
this pull request
Jun 6, 2026
seq 1 2000000 into an 80x24 window took ~31s — 15-25x behind ghostty (~1.5s) and xfce4-terminal (~2.7s). perf blamed upstream, layer by layer: - ultraviolet Buffer.DeleteLineArea shifted cells one struct assignment at a time on every scroll (~80% of CPU) - vt Scrollback.Push evicted via slices.Delete — an O(10k) memmove per line of output at the default cap (~47% of the remainder) - the trailing-blank trim paid interface-unwrapping color compares per cell, and whole-line clears ran wide-cell repair per cell Fixes are upstream PRs (charmbracelet/ultraviolet#119, charmbracelet/x#888; see also #887) — go.mod pins both modules to the LXXero fork commits carrying them until they merge, then the replaces drop. Result: ~6-10s wall, 3-5x faster, within ~3x of the purpose-built terminals from ~20x behind. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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.
Problem
Buffer.DeleteLineArea(the screen-scroll path — every newline at the bottom of the scroll region) shifts cells one struct assignment at a time: O(rows×cols) Cell copies per scrolled line. Profiling a real terminal emulator (xerotty) floodingseq 1 2000000into an 80×24 screen, this single function + the memmoves it generates was ~80% of total process CPU (perf, Linux). At ~100 bytes per Cell that's hundreds of gigabytes moved to scroll two million lines.Change
InsertLineArea(scroll down).Line.Set's wide-cell repair has nothing to repair when every cell in the line is replaced.copy()per row instead of per-cell assignment.Results
End-to-end
seq 1 2000000wall time in xerotty: 31s → 13.5s from this change alone (→ ~6–8s combined with a scrollback ring buffer in x/vt, PR incoming there).Existing tests pass; behavior is identical — lines land in the same places with the same contents, only the storage shuffling changed.