fix(api): serialize tracer Stop against GetResult

[envestcc]

The race

parseTracer arms a timeout watchdog that calls tracer.Stop from its own goroutine, while the goroutine running the trace calls GetResult. geth's StructLogger stores the interruption reason on a plain field:

func (l *StructLogger) Stop(err error) {
	l.reason = err            // plain write
	l.interrupt.Store(true)   // atomic, but *after*
}

func (l *StructLogger) GetResult() (json.RawMessage, error) {
	if l.reason != nil {      // plain read, never consults l.interrupt
		return nil, l.reason
	}
	...

reason is written and read without synchronisation, and the atomic interrupt establishes no happens-before edge for it because GetResult never reads it. Any debug_trace* request that hits its deadline can trip this.

The fix

Wrap the tracer so Stop and GetResult hold a common mutex. Hooks are passed through untouched — they run on the tracing goroutine and coordinate through l.interrupt, which is already atomic. Nothing in the geth fork is modified.

The guard is applied to every branch of parseTracer (default struct logger, configured struct logger, and named JS/native tracers), since the watchdog covers all of them.

Reproduction

TestParseTracerDefaultLoggerTimeoutFires already exercises the path; the race detector is what surfaces it, so no new test is needed.

On master (85fd452):

$ go test -count=1 -gcflags="all=-N -l" -short -race \
    -run TestParseTracerDefaultLoggerTimeoutFires ./api/
exit=1   WARNING: DATA RACE x2

With this commit:

exit=0   WARNING: DATA RACE x0

Full ./api/ package under the same flags: ok, 0 races.

Notes

Found while integrating this fix into rc_2.5.0; opening it separately against master so the branches don't diverge on it.

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud