mirror of
https://github.com/MobileGL-Dev/MobileGL
synced 2026-11-07 10:03:23 +09:00
swung0x48
5fb54f03fe
The spawn work had the two roles append to one MOBILEGL_LOG_FILE_PATH under O_APPEND. It was correct - two processes at one offset would corrupt the file, O_APPEND made each write atomic - but it put two sessions' lines in one place and made every per-side assertion a search. This splits the sink: the client role writes <base>.client.log and the server role writes <base>.server.log. THE BASE NAME NAMES NO FILE ANY MORE, AND THAT IS THE POINT. Leaving the client at the old unsuffixed path would have been the compatible choice and the wrong one: a reader that was not updated would go on finding a file, go on passing, and go on seeing only half the session - and a refusal census that reports a confident zero for the half it cannot see is worse than one that fails. With both names moved, an un-updated reader gets "file not found" and says so. There were 34 such readers; every one is moved here. THE ROLE IS PER-THREAD, not per-process, because under inproc both roles are threads of one process. Log.cpp keys the sink on a thread-local role: the apply thread declares itself server (ServerLoop.cpp), a spawned image is server for its whole life (MOBILEGL_IPC_ROLE=server, read once), and everything else is the client. The naming rule (RoleLogPath) and the two whole-log operations (ReadRoleLogs / TruncateRoleLogs) are EXPORTED from the library, so no reader reimplements the derivation - a second copy is the copy that falls behind. Readers, by how they read: - unit mains set MOBILEGL_LOG_FILE_PATH and read a role file: they keep the base in the env and derive the client path from the library helper. - the death-test child runners (ReadLog) now concatenate BOTH roles: a death test asserts the CHILD said something, and the applier's refusals - the very diagnostics these check - are written under the server role. - the offset-based Emit readers derive the client path once after setenv. - PipeStatsWindow reads both, because draws/vbs increment in the backend's PrepareForDraw - the apply thread, the server role - so the summary line is in the server file even under inproc. - run_trace_case.cmake's split census reads both files; the client keeps the transport markers. - split_log_paths.py (and its self-test, and the control stub) derive both role files from the declared base. - the Android script pulls both role files off the device and concatenates them, so every downstream check is unchanged - and it judges a role file by CONTENT, because `run-as test -f` through `exec-out` returned 0 for a missing file on the device and `exec-out` folds cat's own not-found line into stdout. The monolith arm legitimately has no server file, and that is a silent skip rather than a one-line server log. Guarded so a pull build is untouched: RoleLogPath and the role sink are wholly inside MOBILEGL_BUILD_DISAGGREGATED; a non-disaggregated build writes the base path exactly as before. G1 re-measured - 0 symbol diff, .text unchanged at 0xa52203. Verified on device 2f7cbe2e: monolith writes only the client log (no phantom server log), inproc and spawn each write both and pass their transport proofs (spawn: server role in pid 16125). Unit 2353 (one pre-existing environmental failure), integration-split 186/186, integration-spawn 102/102, retrace 6/6, control smoke 12/12, split-log-paths self-test 9/9.
152 lines
9.5 KiB
Bash
152 lines
9.5 KiB
Bash
#!/bin/bash
|
|
# EXIT GATE E2's PICTURE CONTROL: dropping the draws on the wire must redden this retrace, and
|
|
# it must be THIS control's red.
|
|
#
|
|
# WHY THERE IS A SECOND CONTROL ON THIS LANE, and why it is the draws and not the clears.
|
|
#
|
|
# The retrace-split job already carries scripts/ci/retrace_pull_library_control.sh, which swaps a
|
|
# PULL library in and requires run_trace_case.cmake's transport-resolution assertion to catch it.
|
|
# That control proves the lane is running a SPLIT library. It does not prove that the PICTURE came
|
|
# through the wire, and it cannot: OpenRA scores ssim 1.000000 against a monolith library too, so
|
|
# the transport assertion is what reds it and the comparator is never consulted. BRIEF-P5 7 E2
|
|
# names a second, sharper statement - "patch an emitter to drop a record and the SSIM must fall
|
|
# below the threshold" - and until now nothing executed it.
|
|
#
|
|
# c1 shipped MOBILEGL_IPC_E2_DROP_CLEAR for exactly that, and the joint gate ran it and it DID NOT
|
|
# WORK: with the knob armed and its WARN in the library's own log, the OpenRA retrace under inproc
|
|
# still scored ssim=1.000000 with mismatchPixels=0 (joint-v1.md 3, "E2 clear-drop control - NOT
|
|
# VERIFIED"). That is not a broken knob. `apitrace dump` over the 31249 replayed calls counts 30
|
|
# glClear, 30 glXSwapBuffers and 788 glDrawArrays, and the final frame issues its clear at call
|
|
# 30197 and then covers the surface four times over with a terrain layer before the snapshot at
|
|
# 31249. A frame that overdraws every pixel it clears has a picture that does not depend on the
|
|
# clear. Measured here with the dropped-record count published: 29 Clear records dropped, ssim
|
|
# still 1.000000. The knob worked; the observable was invisible.
|
|
#
|
|
# So this control drops every DrawVbo record instead (MOBILEGL_IPC_E2_DROP_DRAW=1): the surface
|
|
# can then only carry the clear colour, and the golden is made of the geometry. Measured on the
|
|
# same head: 758 DrawVbo records dropped, ssim=0.000036, mismatchPixels=295296.
|
|
#
|
|
# WHAT THE CONTROL ASSERTS, and none of the three is the process exit code alone (R-16):
|
|
# 1. a NON-EMPTY selection, counted before the run - `--no-tests=error` turns an empty selection
|
|
# into a non-zero exit, which is how ID-46 finding 8(b) got a control to congratulate itself;
|
|
# 2. the SSIM ACTUALLY FELL: the numbers are parsed out of the retrace's own summary and
|
|
# compared, rather than "ctest was non-zero". A loader failure, a missing fixture, a timeout
|
|
# and a Fatal{ all exit non-zero and none of them is this control's red;
|
|
# 3. the LIBRARY SAID IT DROPPED SOMETHING: its own "E2 control armed ... N records dropped on
|
|
# the wire" line, with N > 0, from the log file the replay wrote. Without this, a knob that
|
|
# was never read - a stale library, a variable that did not reach the process, an emit table
|
|
# that fell through to the driver - would redden the picture for some other reason and pass.
|
|
#
|
|
# Usage: retrace_drop_draw_control.sh <case> <backend>
|
|
# CTEST ctest binary (default: ctest)
|
|
# CONTROL_TMPDIR scratch dir (default: ${RUNNER_TEMP:-/tmp})
|
|
# LIBRARY_LOG the replay's CLIENT library log (default: <case>/<backend>/output/mobilegl.client.log)
|
|
set -u
|
|
|
|
CASE="${1:?usage: retrace_drop_draw_control.sh <case> <backend>}"
|
|
BACKEND="${2:?usage: retrace_drop_draw_control.sh <case> <backend>}"
|
|
|
|
CTEST="${CTEST:-ctest}"
|
|
CONTROL_TMPDIR="${CONTROL_TMPDIR:-${RUNNER_TEMP:-/tmp}}"
|
|
# P6: BOTH ROLES HAVE THEIR OWN LOG and neither keeps the old name, so an un-updated default
|
|
# here would be a file that does not exist rather than half a session read as a whole one. This
|
|
# control's marker (`E2 control armed`) is a CLIENT-side line.
|
|
LIBRARY_LOG="${LIBRARY_LOG:-${CASE}/${BACKEND}/output/mobilegl.client.log}"
|
|
mkdir -p "${CONTROL_TMPDIR}"
|
|
FROZEN_LIBRARY="${FROZEN_LIBRARY:?FROZEN_LIBRARY must name the split library the replay loads}"
|
|
symbols=$(nm --defined-only "${FROZEN_LIBRARY}") || exit 1
|
|
remote_count=$(printf '%s\n' "${symbols}" | grep -ic MG_Remote || true)
|
|
echo "draw-drop control library: ${FROZEN_LIBRARY}: MG_Remote=${remote_count}"
|
|
if [ "${remote_count}" -lt 1 ]; then
|
|
echo '::error::draw-drop control requires a split library: MG_Remote=0'
|
|
exit 1
|
|
fi
|
|
|
|
selector="^MobileGLTraceReplay\.${CASE}\.${BACKEND}$"
|
|
|
|
# The rerun replays into the same case directory, so the good run's images are put aside and
|
|
# restored whichever way the control goes; "Upload actual image" runs `if: always()` and would
|
|
# otherwise ship the deliberately-wrong run's output under the good run's name.
|
|
GOOD_OUTPUT="${CONTROL_TMPDIR}/drop-draw-verified-output"
|
|
rm -rf "${GOOD_OUTPUT}"
|
|
if [ -d "${CASE}" ]; then cp -a "${CASE}" "${GOOD_OUTPUT}"; fi
|
|
|
|
restore_good_output() {
|
|
if [ -d "${GOOD_OUTPUT}" ]; then
|
|
rm -rf "${CASE}"; mv "${GOOD_OUTPUT}" "${CASE}"
|
|
echo "restored the verified run's output over the control's"
|
|
fi
|
|
}
|
|
trap restore_good_output EXIT
|
|
trap 'exit 130' INT TERM
|
|
|
|
matched=$("${CTEST}" -N -R "${selector}" | grep -cE '^ *Test *#[0-9]+:')
|
|
if [ "${matched}" -lt 1 ]; then
|
|
restore_good_output
|
|
echo "::error::the control selected ${matched} tests with -R '${selector}', so there is nothing for the dropped draws to redden. --no-tests=error would have exited non-zero on the empty selection and a control without this guard reads that as success (ID-46 finding 8b)."
|
|
exit 1
|
|
fi
|
|
|
|
# A PREVIOUS RUN'S LINE MUST NEVER ARM THIS ONE. The CLIENT truncates this log once at open and
|
|
# then appends (P6 made the sink O_APPEND, because under spawn a second process writes the same
|
|
# file and two truncating handles overwrite each other's bytes) - but it only truncates if the
|
|
# replay gets that far, and a run that died in the loader would leave the baseline's log in place
|
|
# with a perfectly good "control armed" line in it. Removing it first is the same rule
|
|
# split_negative_controls.sh's `reset` step follows.
|
|
rm -f "${LIBRARY_LOG}"
|
|
|
|
out="${CONTROL_TMPDIR}/retrace-drop-draw-output.txt"
|
|
# THE TRANSPORT COMES FROM THE JOB, not from this file. retrace-split is a matrix over
|
|
# {backend, case, transport} as of P6, and a control that hard-coded `inproc` would have gone on
|
|
# proving something about the OTHER arm while the spawn arm ran unguarded. Default inproc so a
|
|
# caller that sets nothing behaves exactly as before.
|
|
transport="${SPLIT_TRANSPORT:-inproc}"
|
|
export MOBILEGL_TRANSPORT="${transport}"
|
|
export MOBILEGL_IPC_E2_DROP_DRAW=1
|
|
"${CTEST}" -V --no-tests=error --timeout 10800 -R "${selector}" > "${out}" 2>&1
|
|
control_rc=$?
|
|
unset MOBILEGL_IPC_E2_DROP_DRAW
|
|
cat "${out}"
|
|
|
|
# Read the library's evidence BEFORE the good output is restored over it.
|
|
armed_line=""
|
|
dropped=0
|
|
if [ -f "${LIBRARY_LOG}" ]; then
|
|
armed_line=$(grep 'MGPipe: E2 control armed' "${LIBRARY_LOG}" | tail -1)
|
|
dropped=$(printf '%s' "${armed_line}" | sed -n 's/.*armed[^,]*, \([0-9][0-9]*\) records dropped.*/\1/p')
|
|
dropped="${dropped:-0}"
|
|
fi
|
|
cp -f "${LIBRARY_LOG}" "${CONTROL_TMPDIR}/drop-draw-library.log" 2>/dev/null
|
|
|
|
restore_good_output
|
|
|
|
if [ "${control_rc}" -eq 0 ]; then
|
|
echo "::error::the split retrace PASSED with every DrawVbo record dropped on the wire. The golden is made of that geometry, so a green here means the picture did not come from the wire: the emit table fell through to the driver, the library under test is not the one the lane thinks it is, or MOBILEGL_IPC_E2_DROP_DRAW never reached the process. Exit gate E2 is exactly this statement and nothing weaker - MOBILEGL_IPC_E2_DROP_CLEAR is NOT a substitute (measured: 29 clears dropped, ssim still 1.000000, because OpenRA overdraws every pixel it clears)."
|
|
exit 1
|
|
fi
|
|
|
|
# 2. THE SSIM ACTUALLY FELL. Parsed, not inferred from the exit code.
|
|
ssim_line=$(grep -o 'ssim=[0-9.]*, ssimThreshold=[0-9.]*' "${out}" | tail -1)
|
|
if [ -z "${ssim_line}" ]; then
|
|
echo "::error::the split retrace went red (ctest exit ${control_rc}) with the draws dropped, but its output carries no 'ssim=..., ssimThreshold=...' summary at all, so the comparator never ran. A loader failure, a missing fixture, a timeout or a Fatal{ all land here and none of them is this control's red."
|
|
exit 1
|
|
fi
|
|
ssim=${ssim_line#ssim=}; ssim=${ssim%%,*}
|
|
threshold=${ssim_line##*ssimThreshold=}
|
|
if ! awk -v a="${ssim}" -v b="${threshold}" 'BEGIN { exit !(a + 0 < b + 0) }'; then
|
|
echo "::error::the split retrace went red (ctest exit ${control_rc}) but its ${ssim_line} is NOT below the threshold, so the picture is not what reddened it. This control's whole claim is that the golden is made of the dropped geometry."
|
|
exit 1
|
|
fi
|
|
|
|
# 3. THE LIBRARY SAID IT DROPPED SOMETHING.
|
|
if [ -z "${armed_line}" ]; then
|
|
echo "::error::the split retrace went red with ssim ${ssim} < ${threshold}, but ${LIBRARY_LOG} carries no 'MGPipe: E2 control armed' line, so there is no evidence the knob was ever read by the process that produced the picture. A library that is not the one under test, an emit table that fell through to the driver, or an environment that did not reach the replay all produce a wrong picture for a reason that has nothing to do with this control."
|
|
exit 1
|
|
fi
|
|
if [ "${dropped}" -lt 1 ]; then
|
|
echo "::error::the split retrace went red with ssim ${ssim} < ${threshold} and the knob announced itself - '${armed_line}' - but it reports ZERO records dropped. The emitter was never reached, so whatever changed the picture was not this control. This is the R-16 case the dropped-record COUNT exists for: the arming message alone proves only that the knob was read."
|
|
exit 1
|
|
fi
|
|
|
|
echo "the dropped draws turned the split retrace red for their own reason (ctest exit ${control_rc}): ${matched} selected case(s), ssim ${ssim} < ${threshold}, and the library dropped ${dropped} record(s) on the wire"
|