Re: [PATCH 4/4] fsmonitor: Delay updating state until after split index is merged

[Jeff King <peff@peff.net>]

On Fri, Oct 20, 2017 at 03:16:20PM +0200, Johannes Schindelin wrote:

> >  void tweak_fsmonitor(struct index_state *istate)
> >  {
> > +	int i;
> > +
> > +	if (istate->fsmonitor_dirty) {
> > +		/* Mark all entries valid */
> > +		trace_printf_key(&trace_fsmonitor, "fsmonitor is enabled; cache is %d", istate->cache_nr);
> 
> Sadly, a call to trace_printf_key() is not really a noop when tracing is
> disabled. The call to trace_printf_key() hands off to trace_vprintf_fl(),
> which in turn calls prepare_trace_line() which asks trace_want() whether
> we need to trace, which finally calls get_trace_fd(). This last function
> initializes a trace key if needed, and this entire call stack takes time.

It seems like we could pretty easily turn noop traces into a trivial
conditional, like:

diff --git a/trace.h b/trace.h
index 179b249c59..c46b92cbde 100644
--- a/trace.h
+++ b/trace.h
@@ -80,8 +80,11 @@ extern void trace_performance_since(uint64_t start, const char *format, ...);
 #define trace_printf(...) \
 	trace_printf_key_fl(TRACE_CONTEXT, __LINE__, NULL, __VA_ARGS__)
 
-#define trace_printf_key(key, ...) \
-	trace_printf_key_fl(TRACE_CONTEXT, __LINE__, key, __VA_ARGS__)
+#define trace_printf_key(key, ...) do { \
+	if (!key->initialized || key->fd) \
+		trace_printf_key_fl(TRACE_CONTEXT, __LINE__, key, __VA_ARGS__) \
+} while(0)
+
 
 #define trace_argv_printf(argv, ...) \
 	trace_argv_printf_fl(TRACE_CONTEXT, __LINE__, argv, __VA_ARGS__)


(OK, that's got an OR, but if we are really pinching instructions we
could obviously store a single "I've been initialized and am disabled"
flag).

I don't have an opinion one way or the other on these particular
messages, but in general I'd like to see _more_ tracing in Git, not
less. I've often traced Git with a debugger or other tools like perf,
but there's real value in the author of code annotating high-level
logical events.

-Peff