git/list[1] front-page[2] threads[3] people[4] search[5] about
 

[PATCH 05/20] read_packed_refs(): use mmap to read the `packed-refs` file

From
Michael Haggerty <mhagger@alum.mit.edu>
Date
Sep 13, 2017, 17:15 UTC
Message-ID
<67c4bf861beb095a63e8db72f1d37d6bfac39cd0.1505319366.git.mhagger@alum.mit.edu>
In-Reply-To
<cover.1505319366.git.mhagger@alum.mit.edu>

It's still done in a pretty stupid way, involving more data copying than necessary. That will improve in future commits.

Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
---
 refs/packed-backend.c | 42 ++++++++++++++++++++++++++++++++----------
 1 file changed, 32 insertions(+), 10 deletions(-)
diff --git a/refs/packed-backend.c b/refs/packed-backend.c
index 5c50c223ef..154abbd83a 100644
--- a/refs/packed-backend.c
+++ b/refs/packed-backend.c
@@ -215,8 +215,12 @@ static NORETURN void die_invalid_line(const char *path,
  */
 static struct packed_ref_cache *read_packed_refs(struct packed_ref_store *refs)
 {
-	FILE *f;
 	struct packed_ref_cache *packed_refs = xcalloc(1, sizeof(*packed_refs));
+	int fd;
+	struct stat st;
+	size_t size;
+	char *buf;
+	const char *pos, *eol, *eof;
 	struct ref_entry *last = NULL;
 	struct strbuf line = STRBUF_INIT;
 	enum { PEELED_NONE, PEELED_TAGS, PEELED_FULLY } peeled = PEELED_NONE;
@@ -227,8 +231,8 @@ static struct packed_ref_cache *read_packed_refs(struct packed_ref_store *refs)
 	packed_refs->cache = create_ref_cache(NULL, NULL);
 	packed_refs->cache->root->flag &= ~REF_INCOMPLETE;
 
-	f = fopen(refs->path, "r");
-	if (!f) {
+	fd = open(refs->path, O_RDONLY);
+	if (fd < 0) {
 		if (errno == ENOENT) {
 			/*
 			 * This is OK; it just means that no
@@ -241,16 +245,27 @@ static struct packed_ref_cache *read_packed_refs(struct packed_ref_store *refs)
 		}
 	}
 
-	stat_validity_update(&packed_refs->validity, fileno(f));
+	stat_validity_update(&packed_refs->validity, fd);
+
+	if (fstat(fd, &st) < 0)
+		die_errno("couldn't stat %s", refs->path);
+
+	size = xsize_t(st.st_size);
+	buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
+	pos = buf;
+	eof = buf + size;
 
 	dir = get_ref_dir(packed_refs->cache->root);
-	while (strbuf_getwholeline(&line, f, '\n') != EOF) {
+	while (pos < eof) {
 		struct object_id oid;
 		const char *refname;
 		const char *traits;
 
-		if (!line.len || line.buf[line.len - 1] != '\n')
-			die_unterminated_line(refs->path, line.buf, line.len);
+		eol = memchr(pos, '\n', eof - pos);
+		if (!eol)
+			die_unterminated_line(refs->path, pos, eof - pos);
+
+		strbuf_add(&line, pos, eol + 1 - pos);
 
 		if (skip_prefix(line.buf, "# pack-refs with:", &traits)) {
 			if (strstr(traits, " fully-peeled "))
@@ -258,7 +273,7 @@ static struct packed_ref_cache *read_packed_refs(struct packed_ref_store *refs)
 			else if (strstr(traits, " peeled "))
 				peeled = PEELED_TAGS;
 			/* perhaps other traits later as well */
-			continue;
+			goto next_line;
 		}
 
 		refname = parse_ref_line(&line, &oid);
@@ -291,11 +306,18 @@ static struct packed_ref_cache *read_packed_refs(struct packed_ref_store *refs)
 		} else {
 			die_invalid_line(refs->path, line.buf, line.len);
 		}
+
+	next_line:
+		/* The "+ 1" is for the LF character. */
+		pos = eol + 1;
+		strbuf_reset(&line);
 	}
 
-	fclose(f);
-	strbuf_release(&line);
+	if (munmap(buf, size))
+		die_errno("error ummapping packed-refs file");
+	close(fd);
 
+	strbuf_release(&line);
 	return packed_refs;
 }
 
-- 
2.14.1
Previous: Michael HaggertyNext: Michael Haggerty
Message 5 of 27 in “Read `packed-refs` using mmap()”
  1. 00/20 Read `packed-refs` using mmap()Michael Haggerty, Sep 13, 2017
  2. 01/20 ref_iterator: keep track of whether the iterator output is orderedMichael Haggerty, Sep 13, 2017
  3. 02/20 prefix_ref_iterator: break when we leave the prefixMichael Haggerty, Sep 13, 2017
  4. 03/20 packed_ref_cache: add a backlink to the associated `packed_ref_store`Michael Haggerty, Sep 13, 2017
  5. 05/20 read_packed_refs(): use mmap to read the `packed-refs` fileMichael Haggerty, Sep 13, 2017
  6. 06/20 read_packed_refs(): only check for a header at the top of the fileMichael Haggerty, Sep 13, 2017
  7. 07/20 read_packed_refs(): make parsing of the header line more robustMichael Haggerty, Sep 13, 2017
  8. 09/20 packed_ref_cache: remember the file-wide peeling stateMichael Haggerty, Sep 13, 2017
  9. 11/20 mmapped_ref_iterator_advance(): no peeled value for broken refsMichael Haggerty, Sep 13, 2017
  10. 12/20 packed_ref_cache: keep the `packed-refs` file open and mmappedMichael Haggerty, Sep 13, 2017
  11. 17/20 packed_ref_store: get rid of the `ref_cache` entirelyMichael Haggerty, Sep 13, 2017
  12. 19/20 mmapped_ref_iterator: inline into `packed_ref_iterator`Michael Haggerty, Sep 13, 2017
  13. 20/20 packed-backend.c: rename a bunch of things and update commentsMichael Haggerty, Sep 13, 2017
  14. Stefan BellerSep 13, 2017
  15. 14/20 packed_ref_iterator_begin(): iterate using `mmapped_ref_iterator`Michael Haggerty, Sep 13, 2017
  16. 18/20 ref_cache: remove support for storing peeled valuesMichael Haggerty, Sep 13, 2017
  17. 10/20 mmapped_ref_iterator: add iterator over a packed-refs fileMichael Haggerty, Sep 13, 2017
  18. 15/20 packed_read_raw_ref(): read the reference from the mmapped bufferMichael Haggerty, Sep 13, 2017
  19. 16/20 ref_store: implement `refs_peel_ref()` genericallyMichael Haggerty, Sep 13, 2017
  20. 04/20 die_unterminated_line(), die_invalid_line(): new functionsMichael Haggerty, Sep 13, 2017
  21. 13/20 read_packed_refs(): ensure that references are ordered when readMichael Haggerty, Sep 13, 2017
  22. 08/20 read_packed_refs(): read references with minimal copyingMichael Haggerty, Sep 13, 2017
  23. Junio C HamanoSep 13, 2017
  24. Michael HaggertySep 14, 2017
  25. Johannes SchindelinSep 14, 2017
  26. Johannes SchindelinSep 14, 2017
  27. Michael HaggertySep 15, 2017

Read the whole thread, see it on lore, or plain text.

$ cat FOOTERMessages come from the public archive at lore.kernel.org/git, fetched every hour. The front page is chosen and written each morning by an AI editor and can be wrong; the threads themselves are the record. About and API. For agents: an MCP server at https://gitlist.dev/mcp, and any thread, story or person page as Markdown by adding .md to its URL (or sending Accept: text/markdown). Details in /llms.txt.