myxce/.wolf/bugs/bug-1930.md

id	status	status_updated	status_note	component	severity	timestamp	last_seen	occurrences	files	tags	related	fix_commits	gitea_issue	gitea_synced_status	gitea_synced_hash
bug-1930	fixed	2026-10-09		scheduler	wrong-result	2026-10-09T03:28:14Z	2026-10-09T03:28:14Z	1	src/scheduler/src/repl_trace.c src/scheduler/src/repl_trace_store.c src/scheduler/src/repl_internal.h src/scheduler/testsuite/test_repl_trace_cmd.c	repl sim trace jobid	bug-1926	ba6f42cd	355	fixed	b18384b47412

bug-1930: xce_scheduler --sim REPL trace on job rejects the raw decimal job id info trace accepts

Description

Observed: trace on job rejects a raw packed decimal job id, the form info trace and info job accept for the same job:

[sim] trace: invalid job id: 104 (invalid job identity: expected <cluster>-<apiserver>-<timestamp>-<sequence> optionally followed by [k] or [a-b] (decimal digits only, no sign or whitespace))

while info trace 104 and info job 104 both resolve job 0-0-0-104. In the unit test the same rejection for a full packed id:

unexpected output for 'trace on job 81064794341244970': [sim] trace: invalid job id: 81064794341244970 (invalid job identity: expected <cluster>-<apiserver>-<timestamp>-<sequence> optionally followed by [k] or [a-b] (decimal digits only, no sign or whitespace))

Expected: trace on job accepts the same forms as info trace/info job -- the dashed id, X[k], X[a-b] and a raw packed decimal id -- and a raw decimal id selects exactly what its dashed bare form selects (for an array-aligned id, the exact job plus the whole array, tasks.md Implementation Notes 4.3). Impact: inconsistent REPL grammar: an operator who copies a raw id (as info job/info trace accept it) into trace on job gets an error and no trace. No scheduling effect. Found by: user report during scheduler-sim-decision-trace validation.

Environment

• OS: no /etc/os-release; Docker container; NixOS kernel; kernel Linux 6.18.33 x86_64
• libc: glibc 2.42-47 (interpreter /nix/store/wb6rhpznjfczwlwx23zmdrrw74bayxw4-glibc-2.42-47/lib/ld-linux-x86-64.so.2)
• Toolchain: gcc (GCC) 15.2.0, meson 1.9.1
• Build: asan (xce.build), default ASAN_OPTIONS; the symptom does not depend on sanitizer options
• Code: scheduler-framework @ 5ca59dc9-dirty
• Deployment: standalone simulator only (xce_scheduler --sim ... --interactive), no controller, no workers; single dev host

Reproduction

1. Build: bash src/tools/xce-build.sh scheduler --jobs 12 --incremental at the Environment commit.
2. printf 'trace on job 104\ninfo trace 104\nquit\n' | TERM=dumb ASAN_OPTIONS=detect_leaks=0 src/xce.build/scheduler/xce_scheduler --sim src/scheduler/testsuite/fixtures/repl_trace_baseline/contention.replay --interactive --logfile <tmp>/sched.log (repo root) -- trace on job 104 prints the "invalid job id" error above; info trace 104 accepts the id.
3. Smaller: test_on_job_raw_decimal_ids in src/scheduler/testsuite/test_repl_trace_cmd.c.

• Frequency: deterministic (every run, 2026-10-09)
• Evidence: /home/claude/tmp/bug-1930/ -- trace-cmd-red.txt (unit test before the fix), trace-cmd-green.txt, live-green.txt (trace on job 104 after the fix, tracing 0-0-0-104 through continue and info trace 104)

Root cause

parse_sel() in src/scheduler/src/repl_trace.c:210 (before the fix) passes each JOBSEL word straight to libxce's parse_jobsel(), whose grammar is only the dashed cluster-apiserver-timestamp-sequence id with an optional [k]/[a-b] (utils.h), and reports any error. info trace's info_trace_sel() (src/scheduler/src/repl_trace_store.c:456) adds a fallback after a failed parse_jobsel(): an all-digits argument is taken as a raw packed id (strtoull), matching info job. The two selector parsers were written separately and only info trace got the fallback.

Since: aa4c43c8 (scheduler-sim-decision-trace task 4.3, the trace command parser).

Scope: break ... jobid= has the opposite problem (it takes only the raw form) and is tracked separately as bug-1926; not changed here.

Fix

• src/scheduler/src/repl_trace_store.c: new repl_parse_jobsel_arg(arg, out, raw) (declared in repl_internal.h): parse_jobsel() plus the raw packed decimal fallback, returned as JOBSEL_BARE with key = the number and *raw = true; a decimal too wide for 64 bits is JOBSEL_EWIDTH (before, info trace passed it to strtoull, which saturated to UINT64_MAX). info_trace_sel() uses it and keeps its existing behaviour (a raw id names that job).
• src/scheduler/src/repl_trace.c parse_sel(): uses repl_parse_jobsel_arg(). A raw id comes back as the same JOBSEL_BARE selector its dashed form produces, so the existing bare-id rule applies unchanged: the exact selector, plus the whole-array selector when the id is array-aligned (Implementation Notes 4.3). repl_trace.c is 652 lines.
• src/scheduler/testsuite/test_repl_trace_cmd.c: new test_on_job_raw_decimal_ids; the old "7 is rejected" case now uses 7x (a raw 7 is a valid id now), and an over-wide decimal is rejected with the width error.

Verification

• Test: test_on_job_raw_decimal_ids in src/scheduler/testsuite/test_repl_trace_cmd.c (raw plain id, raw array-aligned id and 104 each give the same selectors as their dashed form; mixed raw/dashed/range line) and the over-wide case in test_invalid_arguments_leave_settings: RED before the fix (test_on_job_raw_decimal_ids: FAIL ... unexpected output for 'trace on job 81064794341244970': [sim] trace: invalid job id ...; the width case got the syntax error instead), GREEN after (14 Tests 0 Failures, leak detection on).
• Suites: bash src/tools/xce-build.sh scheduler --jobs 12 --incremental: scheduler suite 118/118 OK, asan (log: /home/claude/tmp/bug-1930/gate-scheduler-suite.log).
• Live: Reproduction step 2 after the fix: [sim] trace: enabled; applies from the next session, then [trace] job 0-0-0-104 ... blocks for sessions 1-5 under continue, and info trace 104 shows the stored trace (live-green.txt).

References

• Related: bug-1926 (break ... jobid= accepts only the raw form; separate fix)
• Docs: .kiro/specs/scheduler-sim-decision-trace/tasks.md (Implementation Notes 4.3), requirements.md 1.2-1.3