git@vger.kernel.org mailing list mirror (one of many)
 help / color / mirror / code / Atom feed
From: Victoria Dye <vdye@github.com>
To: Elijah Newren <newren@gmail.com>, git@vger.kernel.org
Cc: Derrick Stolee <stolee@gmail.com>,
	Lessley Dennington <lessleydennington@gmail.com>
Subject: Re: [RFC PATCH 5/5] Accelerate ensure_skip_worktree_means_skip_worktree by caching
Date: Tue, 11 Jan 2022 13:30:13 -0500	[thread overview]
Message-ID: <950da537-941f-dfb4-460a-1aa4c68e0750@github.com> (raw)
In-Reply-To: <20220109045732.2497526-6-newren@gmail.com>

Elijah Newren wrote:
> Rather than lstat()'ing every SKIP_WORKTREE path, take advantage of the
> fact that entire directories will often be missing, especially for cone
> mode and even more so ever since commit 55dfcf9591 ("sparse-checkout:
> clear tracked sparse dirs", 2021-09-08).  If we have already determined
> that the parent directory of a file (or any other previous ancestor)
> does not exist, then we already know the file cannot exist and do not
> need to lstat() it separately.
> 
> Granted, the cost of ensure_skip_worktree_means_skip_worktree() might
> be considered a bit high for non-cone mode since it might now lstat()
> every SKIP_WORKTREE path when the index is loaded (an O(N) cost, with
> N the number of SKIP_WORKTREE paths), but non-cone mode users already
> have to deal with the O(N*M) cost (with N=the number of tracked files
> and M=the number of sparsity patterns), so this should be reasonable.
> 

Did you write/run any performance tests to see how this optimization changed
the execution time? If not, running the `p2000` performance tests against
the patch series base, [3/5], and [5/5] would provide some really helpful
insight into the cost of `ensure_skip_worktree_means_skip_worktree`, then
how much this optimization improves it.

> Signed-off-by: Elijah Newren <newren@gmail.com>
> ---
>  sparse-index.c | 105 ++++++++++++++++++++++++++++++++++++++++++++++++-
>  1 file changed, 103 insertions(+), 2 deletions(-)
> 
> diff --git a/sparse-index.c b/sparse-index.c
> index 79d50e444c..608782e255 100644
> --- a/sparse-index.c
> +++ b/sparse-index.c
> @@ -341,18 +341,117 @@ void ensure_correct_sparsity(struct index_state *istate)
>  		ensure_full_index(istate);
>  }
>  
> +struct path_cache_entry {
> +	struct hashmap_entry ent;
> +	const char *path;
> +	int path_length;
> +	int is_present;
> +};
> +
> +static int path_cache_cmp(const void *unused,
> +			  const struct hashmap_entry *entry1,
> +			  const struct hashmap_entry *entry2,
> +			  const void *also_unused)
> +{
> +	const struct path_cache_entry *e1, *e2;
> +
> +	e1 = container_of(entry1, const struct path_cache_entry, ent);
> +	e2 = container_of(entry2, const struct path_cache_entry, ent);
> +	if (e1->path_length != e2->path_length)
> +		return e1->path_length - e2->path_length;
> +	return memcmp(e1->path, e2->path, e1->path_length);
> +}
> +
> +static struct path_cache_entry *find_path_cache_entry(struct hashmap *map,
> +						      const char *str,
> +						      int str_length)
> +{
> +	struct path_cache_entry entry;
> +	hashmap_entry_init(&entry.ent, memhash(str, str_length));
> +	entry.path = str;
> +	entry.path_length = str_length;
> +	return hashmap_get_entry(map, &entry, ent, NULL);
> +}
> +
> +static void record(struct hashmap *path_cache,
> +		   struct mem_pool *pool,
> +		   const char *path,
> +		   int path_length,
> +		   int found)
> +{
> +	struct path_cache_entry *entry;
> +
> +	entry = mem_pool_alloc(pool, sizeof(*entry));
> +	hashmap_entry_init(&entry->ent, memhash(path, path_length));
> +	entry->path = path;
> +	entry->path_length = path_length;
> +	entry->is_present = found;
> +	hashmap_add(path_cache, &entry->ent);
> +}
> +
> +static int path_found(struct hashmap *path_cache, struct mem_pool *pool,
> +		      const char *path, int path_length)
> +{
> +	struct stat st;
> +	int found;
> +	const char *dirsep = path + path_length - 1;
> +	const char *tmp;
> +
> +	/* Find directory separator; memrchr is sadly glibc-specific */
> +	while (dirsep > path && *dirsep != '/')
> +		dirsep--;
> +
> +	/* If parent of path doesn't exist, no point lstat'ing path... */
> +	if (dirsep > path) {
> +		struct path_cache_entry *entry;
> +		int new_length, parent_found;
> +
> +		/* First, check if path's parent's existence was cached */
> +		new_length = dirsep - path;
> +		entry = find_path_cache_entry(path_cache, path, new_length);
> +		if (entry)
> +			parent_found = entry->is_present;
> +		else
> +			parent_found = path_found(path_cache, pool,
> +						  path, new_length);
> +
> +		if (!parent_found) {
> +			/* path can't exist if parent dir doesn't */
> +			record(path_cache, pool, path, path_length, 0);
> +			return 0;
> +		} /* else parent was found so must check path itself too... */
> +	}
> +
> +	/* Okay, parent dir exists, so we have to check original path */
> +
> +	/* Make sure we have a NUL-terminated string to pass to lstat */
> +	tmp = path;
> +	if (path[path_length])
> +		tmp = mem_pool_strndup(pool, path, path_length);
> +	/* Determine if path exists */
> +	found = !lstat(tmp, &st);
> +
> +	record(path_cache, pool, path, path_length, found);
> +	return found;
> +}
> +
>  void ensure_skip_worktree_means_skip_worktree(struct index_state *istate)
>  {
> +	struct hashmap path_cache = HASHMAP_INIT(path_cache_cmp, NULL);
> +	struct mem_pool pool;
> +
>  	int i;
> +
>  	if (!core_apply_sparse_checkout)
>  		return;
>  
> +	mem_pool_init(&pool, 32*1024);
>  restart:
>  	for (i = 0; i < istate->cache_nr; i++) {
>  		struct cache_entry *ce = istate->cache[i];
> -		struct stat st;
>  
> -		if (ce_skip_worktree(ce) && !lstat(ce->name, &st)) {
> +		if (ce_skip_worktree(ce) &&
> +		    path_found(&path_cache, &pool, ce->name, strlen(ce->name))) {
>  			if (S_ISSPARSEDIR(ce->ce_mode)) {
>  				ensure_full_index(istate);
>  				goto restart;
> @@ -360,6 +459,8 @@ void ensure_skip_worktree_means_skip_worktree(struct index_state *istate)
>  			ce->ce_flags &= ~CE_SKIP_WORKTREE;
>  		}
>  	}
> +	hashmap_clear(&path_cache);
> +	mem_pool_discard(&pool, 0);
>  }
>  
>  


  reply	other threads:[~2022-01-11 18:30 UTC|newest]

Thread overview: 11+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2022-01-09  4:57 [RFC PATCH 0/5] Remove the present-despite-SKIP_WORKTREE class of bugs Elijah Newren
2022-01-09  4:57 ` [RFC PATCH 1/5] t1011: add testcase demonstrating accidental loss of user modifications Elijah Newren
2022-01-09  4:57 ` [RFC PATCH 2/5] unpack-trees: fix accidental loss of user changes Elijah Newren
2022-01-09  4:57 ` [RFC PATCH 3/5] repo_read_index: ensure SKIP_WORKTREE means skip worktree Elijah Newren
2022-01-10 20:38   ` Victoria Dye
2022-01-11 19:27     ` Elijah Newren
2022-01-11 23:09       ` Victoria Dye
2022-01-09  4:57 ` [RFC PATCH 4/5] Update documentation related to sparsity and the skip-worktree bit Elijah Newren
2022-01-09  4:57 ` [RFC PATCH 5/5] Accelerate ensure_skip_worktree_means_skip_worktree by caching Elijah Newren
2022-01-11 18:30   ` Victoria Dye [this message]
2022-01-11 22:04     ` Elijah Newren

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

  List information: http://vger.kernel.org/majordomo-info.html

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=950da537-941f-dfb4-460a-1aa4c68e0750@github.com \
    --to=vdye@github.com \
    --cc=git@vger.kernel.org \
    --cc=lessleydennington@gmail.com \
    --cc=newren@gmail.com \
    --cc=stolee@gmail.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
Code repositories for project(s) associated with this public inbox

	https://80x24.org/mirrors/git.git

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).