I am very enticed by the subject!

On 1/29/2020 5:03 PM, Elijah Newren via GitGitGadget wrote:

Unfortunately, commit df5bcdf83aeb ("dir: recurse into untracked dirs
for ignored files", 2017-05-18), added exactly such a case to the code,

I was disappointed that the commit you mention did not add a test for
the new behavior, but then found a test change in the following commit
fb89888849 (dir: hide untracked contents of untracked dirs, 2017-05-18).
This makes me feel better that your changes are less likely to un-do
the intention of df5bcdf83aeb.

meaning we'd have two calls to read_directory_recursive() for an
untracked directory.  So, if we had a file named
   one/two/three/four/five/somefile.txt
and nothing in one/ was tracked, then 'git status --ignored' would
call read_directory_recursive() twice on the directory 'one/', and
each of those would call read_directory_recursive() twice on the
directory 'one/two/', and so on until read_directory_recursive() was
called 2^5 times for 'one/two/three/four/five/'.

Wow! Good find. "Accidentally exponential" is a lot worse than
"accidentally quadratic". At least the N here _usually_ does not
grow too quickly, but the constant here (lstat-ing directories and
files) is significant enough that 2^3 or 2^4 is enough to notice
the difference.

Avoid calling read_directory_recursive() twice per level by moving a
lot of the special logic into treat_directory().

Since dir.c is somewhat complex, extra cruft built up around this over
time.  While trying to unravel it, I noticed several instances where the
first call to read_directory_recursive() would return e.g.
path_untracked for a some directory and a later one would return e.g.
path_none, and the code relied on the side-effect of the first adding
untracked entries to dir->entries in order to get the correct output
despite the supposed override in return value by the later call.

I am somewhat concerned that there are still bugs and maybe even
testcases with the wrong expectation.  I have tried to carefully
document treat_directory() since it becomes more complex after this
change (though much of this complexity came from elsewhere that probably
deserved better comments to begin with).  However, much of my work felt
more like a game of whackamole while attempting to make the code match
the existing regression tests than an attempt to create an
implementation that matched some clear design.  That seems wrong to me,
but the rules of existing behavior had so many special cases that I had
a hard time coming up with some overarching rules about what correct
behavior is for all cases, forcing me to hope that the regression tests
are correct and sufficient.  (I'll note that this turmoil makes working
with dir.c extremely unpleasant for me; I keep hoping it'll get better,
but it never seems to.)

Keep fighting the good fight! It appears that some of our most-important
code has these complicated cases and side-effects because it has grown
so organically over time. It's unlikely that someone _could_ rewrite it
to avoid that pain, as dir.c contains a lot of accumulated knowledge from
the many special-cases Git handles. I suppose the only thing we can do
is try to write as many detailed tests as possible.

However, on the positive side, it does make the code much faster.  For
the following simple shell loop in an empty repository:

  for depth in $(seq 10 25)
  do
    dirs=$(for i in $(seq 1 $depth) ; do printf 'dir/' ; done)
    rm -rf dir
    mkdir -p $dirs
    >$dirs/untracked-file
    /usr/bin/time --format="$depth: %e" git status --ignored >/dev/null
  done

I saw the following timings, in seconds (note that the numbers are a
little noisy from run-to-run, but the trend is very clear with every
run):

    10: 0.03
    11: 0.05
    12: 0.08
    13: 0.19
    14: 0.29
    15: 0.50
    16: 1.05
    17: 2.11
    18: 4.11
    19: 8.60
    20: 17.55
    21: 33.87
    22: 68.71
    23: 140.05
    24: 274.45
    25: 551.15

Are these timings on Linux? I imagine that the timings will increase
much more quickly on Windows.

After this fix, those drop to:

    10: 0.00

...

    25: 0.00

Nice. I wonder if presenting these 0.00 values as a table is worth
the space? At least the effect is dramatic.

In fact, it isn't until a depth of 190 nested directories that it
sometimes starts reporting a time of 0.01 seconds and doesn't
consistently report 0.01 seconds until there are 240 nested directories.
The previous code would have taken
  17.55 * 2^220 / (60*60*24*365) = 9.4 * 10^59 YEARS
to have completed the 240 nested directories case.  It's not often
that you get to speed something up by a factor of 3*10^69.

Awesome.

WARNING: This change breaks t7063.  I don't know whether that is to be expected
(I now intentionally visit untracked directories differently so naturally the
untracked cache should change), or if I've broken something.  I'm hoping to get
an untracked cache expert to chime in...

I suppose that when the untracked cache is enabled, your expectation that we
do not need to recurse into an untracked directory is incorrect: we actually
want to explore that directory. Is there a mode we can check to see if we
are REALLY REALLY collecting _all_ untracked paths? Perhaps we need to create
one?

I'm CC'ing Kevin Willford because he is more familiar with the Git index
than me, and perhaps the untracked cache in particular.

Perhaps here is where you could also have a DIR_LIST_ALL_UNTRACKED
flag to ensure the untracked cache loads all untracked paths?

+	 */
+	if (!(dir->flags & (DIR_SHOW_IGNORED_TOO | DIR_HIDE_EMPTY_DIRECTORIES)))
 		return excluded ? path_excluded : path_untracked;
 
+	/*
+	 * If we only want to determine if dirname is empty, then we can
+	 * stop at the first file we find underneath that directory rather
+	 * than continuing to recurse beyond it.  If DIR_SHOW_IGNORED_TOO
+	 * is set, then we want MORE than just determining if dirname is
+	 * empty.
+	 */
+	stop_early = ((dir->flags & DIR_HIDE_EMPTY_DIRECTORIES) &&
+		      !(dir->flags & DIR_SHOW_IGNORED_TOO));
+
+	/*
+	 * If /every/ file within an untracked directory is ignored, then
+	 * we want to treat the directory as ignored (for e.g. status
+	 * --porcelain), without listing the individual ignored files
+	 * underneath.  To do so, we'll save the current ignored_nr, and
+	 * pop all the ones added after it if it turns out the entire
+	 * directory is ignored.

Here is a question for an untracked cache expert: Do we store ignored
paths in the untracked cache?

+	 */
+	old_ignored_nr = dir->ignored_nr;
+
+	/* Actually recurse into dirname now, we'll fixup the state later. */
 	untracked = lookup_untracked(dir->untracked, untracked,
 				     dirname + baselen, len - baselen);
+	state = read_directory_recursive(dir, istate, dirname, len, untracked,
+					 stop_early, stop_early, pathspec);
+
+	/* There are a variety of reasons we may need to fixup the state... */
+	if (state == path_excluded) {
+		int i;
+
+		/*
+		 * When stop_early is set, read_directory_recursive() will
+		 * never return path_untracked regardless of whether
+		 * underlying paths were untracked or ignored (because
+		 * returning early means it excluded some paths, or
+		 * something like that -- see commit 5aaa7fd39aaf ("Improve
+		 * performance of git status --ignored", 2017-09-18)).
+		 * However, we're not really concerned with the status of
+		 * files under the directory, we just wanted to know
+		 * whether the directory was empty (state == path_none) or
+		 * not (state == path_excluded), and if not, we'd return
+		 * our original status based on whether the untracked
+		 * directory matched an exclusion pattern.
+		 */
+		if (stop_early)
+			state = excluded ? path_excluded : path_untracked;
+
+		else {
+			/*
+			 * When
+			 *     !stop_early && state == path_excluded
+			 * then all paths under dirname were ignored.  For
+			 * this case, git status --porcelain wants to just
+			 * list the directory itself as ignored and not
+			 * list the individual paths underneath.  Remove
+			 * the individual paths underneath.
+			 */
+			for (i = old_ignored_nr + 1; i<dir->ignored_nr; ++i)
+				free(dir->ignored[i]);
+			dir->ignored_nr = old_ignored_nr;
+		}
+	}
 
 	/*
-	 * If this is an excluded directory, then we only need to check if
-	 * the directory contains any files.
+	 * If there is nothing under the current directory and we are not
+	 * hiding empty directories, then we need to report on the
+	 * untracked or ignored status of the directory itself.
 	 */
-	return read_directory_recursive(dir, istate, dirname, len,
-					untracked, 1, excluded, pathspec);
+	if (state == path_none && !(dir->flags & DIR_HIDE_EMPTY_DIRECTORIES))
+		state = excluded ? path_excluded : path_untracked;
+
+	/*
+	 * We can recurse into untracked directories that don't match any
+	 * of the given pathspecs when some file underneath the directory
+	 * might match one of the pathspecs.  If so, we should make sure
+	 * to note that the directory itself did not match.
+	 */
+	if (pathspec &&
+	    !match_pathspec(istate, pathspec, dirname, len,
+			    0 /* prefix */, NULL,
+			    0 /* do NOT special case dirs */))
+		state = path_none;
+
+	return state;
 }

This is certainly a substantial change, and I'm not able to read it
carefully right now. I hope to return to it soon, but hopefully I've
pointed out some places that may lead you to resolve your untracked
cache issues.

Thanks,
-Stolee