@gdb — sharing a bounded, read-only evidence packet for gdb/tee-output#3.
Original issue:
gdb/tee-output#3
Exact coordinate
source: gdb/tee-output@c41f8ff383200320b746e953e92709ae1b505a71
environment: GitHub-hosted Linux, Python 3.11
observer: read-only isolated child processes
trajectory: redirect → direct fd writes → immediate close → file verification
Result
Twenty isolated baseline rounds were executed. Each round wrote 64 unique records directly to stdout and 64 to stderr, then called Tee.close() immediately.
baseline failures: 0/20
stdout missing records: 0
stderr missing records: 0
combined missing records: 0
Ten counterfactual rounds used the same exact source and records, changing only shutdown ordering:
flush wrappers
→ restore fd 1/fd 2
→ close writers to deliver EOF
→ wait naturally
→ terminate only after timeout
counterfactual failures: 0/10
Evidence-supported classification:
NOT_REPRODUCED_ON_THIS_RUN_STATIC_RISK_REMAINS
This does not disprove the original report. The current exact source still:
- calls
pause() before _drain();
- has no explicit
sys.stdout.flush() / sys.stderr.flush() in close();
- closes the writer and sends SIGINT to the reader before
wait().
The original report used a macOS path and a minimal print + traceback + immediate close trajectory. This bounded run used Linux non-TTY pipes and direct fd writes, so a macOS/PTY/startup-readiness race remains outside the demonstrated coordinate.
Exact evidence
run: 29773637522
attempt: 1
head SHA: 1b1675d48383c607eb2b26f737db93d79ea619fd
execution SHA: 6d5b56e48e54020ae741121693cb10348592a71c
artifact: 8473778255
artifact digest: sha256:f36759b539384b444b427f04213e9f22c453520bf273f2b40824da3e653774bf
Greg result SHA-256: 6490732f329cc05dc5c06657c92676058b756134b6fc599e4d34c38bfda64736
causal result SHA-256: b2c2857ba08a5125b220f63a17207f575da8f5c7cdae2611ead2c2c71b2bcab2
observer trajectory SHA-256: 2cd8ae207a795185b2a7df8d96704462998b587b41eab9f8af450b120a1bd085
Workflow, code, coordinate model, and exact run:
#97
A useful next matrix would execute the issue’s exact minimal program on macOS with TTY and non-TTY stdout, separating startup readiness from shutdown drain. This report provides evidence only and does not claim that all environments are affected or that production data was lost.
@gdb — sharing a bounded, read-only evidence packet for
gdb/tee-output#3.Original issue:
gdb/tee-output#3
Exact coordinate
Result
Twenty isolated baseline rounds were executed. Each round wrote 64 unique records directly to stdout and 64 to stderr, then called
Tee.close()immediately.Ten counterfactual rounds used the same exact source and records, changing only shutdown ordering:
Evidence-supported classification:
This does not disprove the original report. The current exact source still:
pause()before_drain();sys.stdout.flush()/sys.stderr.flush()inclose();wait().The original report used a macOS path and a minimal
print + traceback + immediate closetrajectory. This bounded run used Linux non-TTY pipes and direct fd writes, so a macOS/PTY/startup-readiness race remains outside the demonstrated coordinate.Exact evidence
Workflow, code, coordinate model, and exact run:
#97
A useful next matrix would execute the issue’s exact minimal program on macOS with TTY and non-TTY stdout, separating startup readiness from shutdown drain. This report provides evidence only and does not claim that all environments are affected or that production data was lost.