Re: "There are too many unreachable loose objects" - why don't we run 'git prune' automatically?

On 09 Jun 2017, at 07:27, Jeff King [off-list ref] wrote:

On Thu, Jun 08, 2017 at 02:45:48PM +0200, Lars Schneider wrote:

quoted

I recently ran into "There are too many unreachable loose objects; run 
'git prune' to remove them." after a "Auto packing the repository in 
background for optimum performance." message.

This was introduced with a087cc9 "git-gc --auto: protect ourselves from 
accumulated cruft" but I don't understand the commit message really.

Why don't we call 'git prune' automatically? I though Git would prune
unreachable objects after 90 days by default anyways. Is the warning 
about unreachable objects that are not yet 90 days old?

We _do_ call "git prune", but we do so with whatever configured
expiration time is (by default 2 weeks; the 90-day expiration is for
reflogs).

The problem is that auto-gc kicked in because there were a bunch of
loose objects, but after repacking and running "git prune" there were
still enough loose objects to trigger auto-gc. Which means every command
you run will do an auto-gc that never actually helps.

So you have two options:

 1. Wait until those objects expire (which may be up to 2 weeks,
    depending on how recent they are), at which point your auto-gc will
    finally delete them.

 2. Run "git prune". Without an argument it prunes everything now,
    with no expiration period.

I agree the existing message isn't great. There should probably be a big
advise() block explaining what's going on (and that expert users can
disable).

How about this?

diff --git a/builtin/gc.c b/builtin/gc.c
index c2c61a57bb..12ee212544 100644
--- a/builtin/gc.c
+++ b/builtin/gc.c

@@ -473,9 +473,18 @@ int cmd_gc(int argc, const char **argv, const char *prefix)
 	if (pack_garbage.nr > 0)
 		clean_pack_garbage();
 
-	if (auto_gc && too_many_loose_objects())
-		warning(_("There are too many unreachable loose objects; "
-			"run 'git prune' to remove them."));
+	if (auto_gc && too_many_loose_objects()) {
+		warning(_("Auto packing did not lead to optimal results as the "
+			"repository contains too many unreachable objects."));
+		advice(_("Unreachable objects are Git objects (commits, files, ...) "
+			"that are not referenced by any branch or tag. This might happen "
+			"if you use 'git rebase' or if you delete branches. Auto packing "
+			"only prunes unreachable objects that are older than 2 weeks "
+			"(default, overridable by the config variable 'gc.pruneExpire'). "
+			"Please run 'git prune' to prune all unreachable objects for "
+			"optimal repository performance."));
+	}
 
 	if (!daemonized)
 		unlink(git_path("gc.log"));
- Lars