fix: deliver queued input and confirmations in non-TUI IOProxy - #697
Open
chrisnestrud wants to merge 1 commit into
Open
chrisnestrud wants to merge 1 commit into
chrisnestrud wants to merge 1 commit into
Conversation
Copy link
Copy Markdown
Sorry, something went wrong.
Root cause: IOProxy.get_input only honored a coder's input queue in TUI mode. In non-TUI mode it delegated straight to the base InputOutput terminal prompt, so answers pushed onto a coder's queue by the WebSocket server, the ACP adapter, or AgentService.wake_primary were silently dropped. confirm_ask had the same gap, so queued confirmations were dropped too. Fix: in non-TUI mode, sweep this coder's queue first. If no payload is waiting and a cancellable prompt session is present, race the terminal read against the per-coder queue wake so a push is delivered even while the prompt is open. The global InterruptibleInput/input() fallback stays a plain blocking read (marked with a ponytail: note) because it cannot be cancelled; a push arriving during that read is delivered on the next get_input() cycle. Shape rationale: the wake event in helpers/queues is global, so the race loop re-checks this coder's own queue after every wake rather than trusting a single signal. A payload for a different reader is put back so its own waiter can consume it. Confirmations reuse the same path and share the TUI payload mapping (never/tweak/all/skip/bool) so both modes agree. Tests cover queued text, a push delivered while the prompt is open, a confirmation left for confirm_ask by get_input, and queued true/false confirmations.
chrisnestrud
force-pushed
the
fix/non-tui-queued-input
branch
from
ce830f2 to
e0d50ed
Compare
September 26, 2026 12:55
This branch has not been deployed
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
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
With tui: false, IOProxy.get_input delegated straight to the base
InputOutput terminal prompt and never drained the per-coder queue that
IOProxy.__init__ registers. So input pushed via queues.push_coder_input(...)
from the WebSocket server (cecli/helpers/server/ws_server.py), the ACP adapter
(cecli/helpers/server/acp/protocol.py), or AgentService.wake_primary
(cecli/helpers/agents/service.py) was silently dropped. confirm_ask had the
same gap, dropping queued confirmations. tui: true was unaffected because
TextualInputOutput.get_input drains the queues.
This also contradicted wake_primary's docstring, which assumes the idle primary
blocks in wait_for_input().
Fix
Non-TUI IOProxy.get_input now:
terminal read against the per-coder queue wake, so a push is delivered even
while the prompt is open.
path, which cannot be cancelled (marked with a ponytail: note naming the
ceiling).
Confirmations take the same route and reuse the TUI payload mapping
(never/tweak/all/skip/bool) so both modes agree.
Why this shape
The wake event in cecli/helpers/queues.py is global, so the race loop
re-checks this coder's own queue after each wake instead of trusting a single
signal. A payload meant for a different reader is put back so its own waiter
still consumes it.
Tests
tests/subagents/test_io_proxy.py: queued text returns (text, uuid) and logs
user_input; a push is delivered while the terminal prompt is open; a
confirmation is left for confirm_ask by get_input; queued true/false
confirmations return the expected values.
Verified offline: tests/subagents/test_io_proxy.py 16 passed,
tests/unit/test_queues.py 15 passed; black (--line-length 100 --preview),
isort, pycodestyle, pyflakes, codespell clean; interactive tui=false re-prompt
loop unchanged (no regression).
Related
Documentation follow-up for this change: #698 (documents the non-TUI
input contract; merge after this PR).