The lane opened in the previous commit was scoped to the Vterm Stage 3
acceptance. Measuring it properly shows the problem is much larger and
not vterm-specific.
Comparing cargo test --list under CI's exact flags against the same
flags plus crdt: 3,024 versus 3,288. 264 tests are dark in CI, and the
single worst line is the library itself at 177 -- cargo test --lib
--features crdt is a required local gate that CI has never run. Ten
suites run zero or one test, including gpu_initial_target (#148's entire
acceptance, 1 of 14), gpu_invocation (#141's, 1 of 14), and a37, the
Stage 3 real-daemon/real-PTY/real-wgpu path that #135 built precisely
because a decoded-message fixture would prove none of the three fit
together.
The lane now carries the per-target table, the verified flag combination
for the fix, a two-part fix shape (a crdt leg on the test job, plus the
GPU-requiring suites onto the existing gpu-render job that already has
lavapipe), and an explicit instruction to sort deliberate exclusions
from accidental ones first -- some of the 264 are perf suites that are
ignored by default and belong to their own jobs, while m10_10_perf has
no ignore attribute and no job naming it.
docs/vterm-framing.md gains an as-framed audit section. The arc is
structurally complete and every test named in the Stage 2 verification
map exists, but criterion 22's "without thrash" clause was never pinned
anywhere -- the word appears nowhere in src or tests -- and that clause
describes exactly the defect #166 fixed. Of the nine Stage 3 tests, only
three drive a real daemon, so the six that construct EditorState
directly could never see a dispatcher-loop defect; a31 passes on the
broken tree for that reason. Four of the nine, including a37 and Stage 3
review round 1's own presence regression guard, do not run in CI at all.
The section also records what was not audited: section 11's blanket
claim about deferral safety covers roughly twenty items and none were
spot-checked.
docs/gpu-terminal-input-framing.md scores bet B2 true now that the
reporter has confirmed typing works, and retracts Q#GT5. The bash fixture
behind it does not reproduce in real use and was almost certainly
measuring its own timing rather than a product behaviour; it is marked
retracted rather than deleted so nobody re-derives it from an earlier
revision.
docs/agent-handoff.md section 5 gains the lesson the confirmation cost:
a daemon-side fix is not deployed until the daemon is restarted from a
tree containing it, and rebuilding a binary does nothing to a running
process.
No code changes.
The dispatcher loop applied BOTH terminal-layout syncs to EVERY attached
frontend. A semantic session satisfies both conditions, because it has a
term_sizes entry from AttachRequest and a semantic terminal declaration,
so its PTY was resized twice on every tick forever: the grid arm
installed the TUI placement size, the semantic arm installed the declared
content rectangle, and each arm's own idempotence guard only ever saw the
size the other had just written. The child took a SIGWINCH storm at tick
cadence and the screen reflowed continuously, which is what made typing
into a GPU terminal impossible while output kept flowing.
The grid arm is also the only per-tick controller-liveness release a
semantic frontend gets, so simply skipping it for those frontends trades
one defect for another: a GPU window that switches away from its terminal
would hold the controller forever, and no peer could resize that PTY
again. The semantic arm cannot take over that job, because the
buffer-follow snapshot clears the viewport declaration that would drive
it.
sync_terminal_layout is therefore split into a frontend-kind-neutral half
(panel reconciliation plus controller liveness, which read only views,
windows and the controller) and a grid-only geometry half (TUI placement
plus the resize). The dispatcher runs the neutral half for every attached
frontend once per tick, then exactly one geometry arm per frontend kind.
sync_terminal_layout survives as the composition of both halves, so the
in-process editor loop and LOCAL are unchanged.
The loop body is extracted into sync_terminal_layouts_for_tick, which
makes the grid/semantic exclusivity structural rather than two adjacent
ifs, and lets the tests drive the real loop body instead of a
re-implementation.
The release that fires when a window has no placement stays in the grid
half deliberately: a semantic frontend has no window_placements entry at
all, so moving it into the neutral half would release a GPU session's
controller on every tick.
Bite-verified against two pre-images, because one is not enough here --
the naive guard fixes the storm and introduces the controller leak, so a
single revert would score the fix complete when it is not:
pin main naive guard split
settle (acc 2+3) FAIL pass pass
controller release (acc 6) pass FAIL pass
grid still resizes (acc 5) pass pass pass
Real-path acceptance: a quiet child that counts SIGWINCH reports 144
frames in 4 s and WINCH 1..12 on screen against the pre-fix tree, versus
a settled screen with the fix. Acceptance 4 (input reaches the child and
returns) is a keep-working pin and passes on both sides -- key transport
was never the defect.
No protocol change; stays v20.
Reported symptom: text input in a terminal does not work on the GPU
frontend while the TUI is fine.
Root cause, measured against a real daemon, a real PTY child, and the
real pmacs-gpu attach client: the dispatcher loop applies BOTH
terminal-layout syncs to EVERY attached frontend each tick. A semantic
session satisfies both conditions -- it has a term_sizes entry from
AttachRequest and a semantic terminal declaration -- so the PTY is
resized twice per tick forever, because the grid arm runs the TUI
placement helper that subtracts a modeline the GPU never draws. The
result is a SIGWINCH storm plus screen reflow at tick cadence, which
makes an interactive line editor unusable while child output keeps
flowing. Measured 730 terminal frames in 20 s for a static screen.
Revision 2 answers Q#GT4 from the code rather than deferring it, which
changes the fix from a one-line guard into a split of
sync_terminal_layout: the grid arm is also the only per-tick
controller-liveness release a GPU session gets, and the semantic arm
neither performs it nor can be made to, because the buffer-follow
snapshot clears the viewport declaration that would drive it.
The doc also records three falsified hypotheses, why the Stage 3
real-path acceptance passes on the broken tree, and a COHERENCE.md
section 6 citation error (two replica frontends have two different
optimistic classifiers; the audit names only one).
No code changes in this commit.
Levi Neuwirth