Myeongseon Choi
dca8fb5a9c
All checks were successful
boundary-lint / PR boundary-claim (Lint (push) Has been skipped
boundary-lint / ban-list lint (Lint (push) Successful in 19s
boundary-lint / duplication-deadline (Layer 1/2) (push) Successful in 19s
Release Publish (Gitea session_helper) / verify-release-tag (push) Successful in 17s
Release Publish (Gitea session_helper) / publish-linux-x86_64 (push) Successful in 4m47s
ci / rust debug (push) Successful in 1m47s
ci / rust release (push) Successful in 2m24s
ci / test-health gate (push) Successful in 16s
ci / mutation test (broker) (push) Successful in 1m14s
ci / python (push) Successful in 1m41s
Two follow-ups to v0.7.42 user reports.
1. Terminal: ``zsh:1: permission denied:`` exit 126
---------------------------------------------------
v0.7.42 dropped the ``${SHELL:-/bin/sh}`` fallback assuming sshd
populates ``$SHELL`` in every login shell, but ``ssh -t host cmd``
runs the user's login shell with ``-c`` (NON-login mode); on some
remotes ``$SHELL`` is unset there, so ``exec "$SHELL" -il`` becomes
``exec "" -il`` → ``permission denied:`` exit 126.
Reinstate the fallback via POSIX ``if [ -z "$SHELL" ]; then
SHELL=/bin/sh; fi`` instead of ``${SHELL:-...}`` so the parser-bug
class that produced ``zsh:1: unknown exec flag -/`` in v0.7.31+ is
still avoided.
2. LSP: cross-file goto-def to unhydrated placeholder lands at (0,0)
--------------------------------------------------------------------
When LSP-pyright / rust-analyzer return a definition target whose
local cache copy is still a 0-byte placeholder, Sublime's
``window.open_file(path:42:5, ENCODED_POSITION)`` cannot place the
caret at row 42 col 5 — that row doesn't exist in an empty buffer —
and clamps to ``(0, 0)``. ``_apply_hydrate_result`` then captured
that ``(0, 0)`` selection before revert and restored it after,
overriding whatever position Sublime defers / re-applies once the
buffer has content. Net result: user lands at the file top instead
of the definition.
Skip capture/restore entirely when the pre-revert buffer was empty.
For the empty-pre-revert case the captured selection is always
``(0, 0)`` — restoring it can only override a Sublime-side deferred
placement, never recover the LSP target — so dropping the restore
is at least as good as before and lets any deferred ENCODED_POSITION
take effect.
The non-empty branch (e21b3a4 cross-file caret fix for already-
hydrated buffers) is unchanged.
Tests
-----
Python 1368 pass; Rust 486 pass.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>