Re: [PATCH 2/2] Remove spurious 'no threads support' warnings

Etienne Buira [off-list ref] writes:

Threads count being defaulted to 0 (autodetect), and --disable-pthreads
build checking that thread count==1, there were spurious warnings about
threads being ignored, despite not specified on command line/conf.

Fixes tests 5521 and 5526 that were broken in --disable-pthreads builds
because of those warnings.

Signed-off-by: Etienne Buira <redacted>
---

I am not sure if this is the right fix.

Shouldn't a --threads=0 from the command line (when there is a
pack.threads configuration hardcoding some number to override it)
give a chance to the auto detection codepath to ask online_cpus()
and receive 1 on NO_PTHREADS build to avoid triggering the same
warning you are squelching with this patch?

That is, something like this instead, perhaps?

-- >8 --
Subject: [PATCH] pack-objects: set number of threads before checking and warning

Under NO_PTHREADS build, we warn when delta_search_threads is not
set to 1, because that is the only sensible value on a single
threaded build.

However, the auto detection that kicks in when that variable is set
to 0 (e.g. there is no configuration variable or command line option,
or an explicit --threads=0 is given from the command line to override
the pack.threads configuration to force auto-detection) was not done
before the condition to issue this warning was tested.

Move the auto-detection code and place it at an appropriate spot.

Signed-off-by: Junio C Hamano <redacted>
---
 builtin/pack-objects.c | 6 ++++--
 thread-utils.h         | 4 ++++
 2 files changed, 8 insertions(+), 2 deletions(-)

diff --git a/builtin/pack-objects.c b/builtin/pack-objects.c
index d391934..a715237 100644
--- a/builtin/pack-objects.c
+++ b/builtin/pack-objects.c

@@ -1972,8 +1972,6 @@ static void ll_find_deltas(struct object_entry **list, unsigned list_size,
 
 	init_threaded_search();
 
-	if (!delta_search_threads)	/* --threads=0 means autodetect */
-		delta_search_threads = online_cpus();
 	if (delta_search_threads <= 1) {
 		find_deltas(list, &list_size, window, depth, processed);
 		cleanup_threaded_search();

@@ -2685,6 +2683,10 @@ int cmd_pack_objects(int argc, const char **argv, const char *prefix)
 		pack_compression_level = Z_DEFAULT_COMPRESSION;
 	else if (pack_compression_level < 0 || pack_compression_level > Z_BEST_COMPRESSION)
 		die("bad pack compression level %d", pack_compression_level);
+
+	if (!delta_search_threads)	/* --threads=0 means autodetect */
+		delta_search_threads = online_cpus();
+
 #ifdef NO_PTHREADS
 	if (delta_search_threads != 1)
 		warning("no threads support, ignoring --threads");

diff --git a/thread-utils.h b/thread-utils.h
index 6fb98c3..d9a769d 100644
--- a/thread-utils.h
+++ b/thread-utils.h

@@ -7,5 +7,9 @@
 extern int online_cpus(void);
 extern int init_recursive_mutex(pthread_mutex_t*);
 
+#else
+
+#define online_cpus() 1
+
 #endif
 #endif /* THREAD_COMPAT_H */

-- 
2.1.2-468-g1a77c5b

On Mon, Oct 13, 2014 at 9:54 PM, Junio C Hamano [off-list ref] wrote:

Etienne Buira [off-list ref] writes:

quoted

Threads count being defaulted to 0 (autodetect), and --disable-pthreads
build checking that thread count==1, there were spurious warnings about
threads being ignored, despite not specified on command line/conf.

Fixes tests 5521 and 5526 that were broken in --disable-pthreads builds
because of those warnings.

Signed-off-by: Etienne Buira <redacted>
---

I am not sure if this is the right fix.

Shouldn't a --threads=0 from the command line (when there is a
pack.threads configuration hardcoding some number to override it)
give a chance to the auto detection codepath to ask online_cpus()
and receive 1 on NO_PTHREADS build to avoid triggering the same
warning you are squelching with this patch?

That is, something like this instead, perhaps?

Indeed, your patch is better.