fix(deps): pin skim to 5.1.0 to unblock Windows picker - #3537
fix(deps): pin skim to 5.1.0 to unblock Windows picker#3537worktrunk-bot wants to merge 1 commit into
Conversation
skim 5.3.0 added Kitty keyboard protocol support that pushes PushKeyboardEnhancementFlags unconditionally in execute_enter and propagates the crossterm error as fatal, so the interactive picker aborts on the legacy Windows console with "Keyboard progressive enhancement not implemented for the legacy Windows API". No Windows capability guard exists as of skim 5.3.2. Pin back to the last green line (5.1.0); 5.2.0 changes border defaults and would churn snapshots.
|
The default branch is still red on this — run 29884002936 (commit 89b58e2, the taiki-e/install-action bump) hit the identical Heads up that #3538 (maintainer-authored) lands the same |
|
Superseded by #3538, which merged the identical If you think this was closed in error, let us know and we'll reopen it.
|
Problem
CI on
mainis red on Windows. Every interactive-picker integration test (test_switch_picker_abort_with_escape,test_switch_picker_alt_l_does_not_hscroll, and the rest ofswitch_picker) fails with:Root cause is the skim 5.1.0 → 5.3.1 bump (#3526), one of three dependabot PRs that merged together at 01:46. skim 5.3.0 added Kitty keyboard protocol support (skim-rs/skim
fix: kitty keyboard protocol), which intui/backend.rs::execute_enterunconditionally runs:On the legacy Windows console API, crossterm's
PushKeyboardEnhancementFlagsreturns an error, and skim propagates it with?as fatal rather than degrading. There is nosupports_keyboard_enhancement()capability guard, and it is still absent as of skim 5.3.2 (which only added packaging CI). The last greenmainrun was the docs commit65f02693, before the skim/tokio/taiki-e bumps landed; tokio and taiki-e can't produce a terminal-keyboard failure.Solution
Pin
skimto the last known-green line,=5.1.0, and revertCargo.lockto match (this restores exactly the pre-bump resolution —frizbee 0.10.0,color-eyrere-added). 5.1.0 predates the Kitty-protocol code entirely, so the picker never touches the Windows-unsupported path.5.2.0 also predates the regression but changes skim's border defaults ("collapsed borders by default"), which would churn picker snapshots across all platforms — out of scope for a CI unblock — so the pin holds at the 5.1 line rather than
<5.3. Lift the pin once skim guards the enhancement behind terminal capability detection.Testing
cargo build --bin wt— compiles with skim 5.1.0.cargo test --test integration --features shell-integration-tests -- switch_picker::test_switch_picker_abort_with_escape switch_picker::test_switch_picker_alt_l_does_not_hscroll— both pass on Linux.The failure itself is Windows-only and can't be reproduced in the Linux sandbox, but 5.1.0 was proven green on every platform (including Windows) before the bump, so restoring it returns
mainto that state. Windows CI on this PR is the confirming signal.Automated fix for failed run