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

Re: [PATCH v2 20/24] fast-import: permit reading multiple marks files

From
Junio C Hamano <gitster@pobox.com>
Date
Jun 5, 2020, 16:14 UTC
Message-ID
<xmqqimg5o5fq.fsf@gitster.c.googlers.com>
In-Reply-To
<20200222201749.937983-21-sandals@crustytoothpaste.net>
"brian m. carlson" <sandals@crustytoothpaste.net> writes:
Show 8 quoted lines
> In the future, we'll want to read marks files for submodules as well.
> Refactor the existing code to make it possible to read multiple marks
> files, each into their own marks set.
>
> Signed-off-by: brian m. carlson <sandals@crustytoothpaste.net>
> ---
>  fast-import.c | 32 ++++++++++++++++++--------------
>  1 file changed, 18 insertions(+), 14 deletions(-)

This conversion looks a bit strange. We are working off of 'marks' that is file scope global, and the topmost caller passes the pointer to that single instance down the callchain to this function (among others) in "s", and it all looked no-op to my eye, but this actually updates the global 'marks'.

Show 12 quoted lines
> diff --git a/fast-import.c b/fast-import.c
> index b8b65a801c..b9ecd89699 100644
> --- a/fast-import.c
> +++ b/fast-import.c
> @@ -493,25 +493,24 @@ static char *pool_strdup(const char *s)
>  	return r;
>  }
>  
> -static void insert_mark(uintmax_t idnum, struct object_entry *oe)
> +static void insert_mark(struct mark_set *s, uintmax_t idnum, struct object_entry *oe)
>  {
> -	struct mark_set *s = marks;

This is no longer needed as the caller gave the 'marks' to us in 's'. Up to this line, this step is a no-op patch.

>  	while ((idnum >> s->shift) >= 1024) {
>  		s = mem_pool_calloc(&fi_mem_pool, 1, sizeof(struct mark_set));
Here we overwrite 's', and we lost the pointer the caller gave us!!!
>  		s->shift = marks->shift + 10;
And we still work off of global marks here and
>  		s->data.sets[0] = marks;
>  		marks = s;

try to swap it here. But if our caller is also using 's' that was passed from its caller, this assignment would not be seen by our caller. We'd need to pass a pointer to caller's 's' to this function and update it around here.

Show 12 quoted lines
>  	}
>  	while (s->shift) {
>  		uintmax_t i = idnum >> s->shift;
>  		idnum -= i << s->shift;
>  		if (!s->data.sets[i]) {
>  			s->data.sets[i] = mem_pool_calloc(&fi_mem_pool, 1, sizeof(struct mark_set));
>  			s->data.sets[i]->shift = s->shift - 10;
>  		}
>  		s = s->data.sets[i];
>  	}
>  	if (!s->data.marked[idnum])
>  		marks_set_count++;

I am not sure if this global also needs a separate counter and I didn't bother to check.

>  	s->data.marked[idnum] = oe;
>  }
With this patch t9300 still seems to pass.

I do not know if this is the fix for Billes Tibor's "fast-import oom" issue <c53bb69b-682d-3b47-4ed0-5f4559e69e37@gmx.com> but the change to this function stood out.

Thanks.
 fast-import.c | 20 +++++++++++---------
 1 file changed, 11 insertions(+), 9 deletions(-)
diff --git a/fast-import.c b/fast-import.c
index 0dfa14dc8c..f39b7890ac 100644
--- a/fast-import.c
+++ b/fast-import.c
@@ -534,13 +534,15 @@ static char *pool_strdup(const char *s)
 	return r;
 }
 
-static void insert_mark(struct mark_set *s, uintmax_t idnum, struct object_entry *oe)
+static void insert_mark(struct mark_set **sp, uintmax_t idnum, struct object_entry *oe)
 {
+	struct mark_set *s = *sp;
+
 	while ((idnum >> s->shift) >= 1024) {
 		s = mem_pool_calloc(&fi_mem_pool, 1, sizeof(struct mark_set));
-		s->shift = marks->shift + 10;
-		s->data.sets[0] = marks;
-		marks = s;
+		s->shift = (*sp)->shift + 10;
+		s->data.sets[0] = (*sp);
+		(*sp) = s;
 	}
 	while (s->shift) {
 		uintmax_t i = idnum >> s->shift;
@@ -958,7 +960,7 @@ static int store_object(
 
 	e = insert_object(&oid);
 	if (mark)
-		insert_mark(marks, mark, e);
+		insert_mark(&marks, mark, e);
 	if (e->idx.offset) {
 		duplicate_count_by_type[type]++;
 		return 1;
@@ -1156,7 +1158,7 @@ static void stream_blob(uintmax_t len, struct object_id *oidout, uintmax_t mark)
 	e = insert_object(&oid);
 
 	if (mark)
-		insert_mark(marks, mark, e);
+		insert_mark(&marks, mark, e);
 
 	if (e->idx.offset) {
 		duplicate_count_by_type[OBJ_BLOB]++;
@@ -1745,12 +1747,12 @@ static void insert_object_entry(struct mark_set *s, struct object_id *oid, uintm
 		e->pack_id = MAX_PACK_ID;
 		e->idx.offset = 1; /* just not zero! */
 	}
-	insert_mark(s, mark, e);
+	insert_mark(&s, mark, e);
 }
 
 static void insert_oid_entry(struct mark_set *s, struct object_id *oid, uintmax_t mark)
 {
-	insert_mark(s, mark, xmemdupz(oid, sizeof(*oid)));
+	insert_mark(&s, mark, xmemdupz(oid, sizeof(*oid)));
 }
 
 static void read_mark_file(struct mark_set *s, FILE *f, mark_set_inserter_t inserter)
@@ -3242,7 +3244,7 @@ static void parse_alias(void)
 		die(_("Expected 'to' command, got %s"), command_buf.buf);
 	e = find_object(&b.oid);
 	assert(e);
-	insert_mark(marks, next_mark, e);
+	insert_mark(&marks, next_mark, e);
 }
 
 static char* make_fast_import_path(const char *path)
Previous: brian m. carlsonNext: Junio C Hamano
Message 38 of 51 in “SHA-256 stage 4 implementation, part 1/3”
  1. 00/24 SHA-256 stage 4 implementation, part 1/3brian m. carlson, Feb 22, 2020
  2. 01/24 builtin/pack-objects: make hash agnosticbrian m. carlson, Feb 22, 2020
  3. Jeff KingFeb 23, 2020
  4. Jeff KingFeb 24, 2020
  5. brian m. carlsonFeb 24, 2020
  6. Jeff KingFeb 24, 2020
  7. 03/24 hex: introduce parsing variants taking hash algorithmsbrian m. carlson, Feb 22, 2020
  8. 07/24 t6300: abstract away SHA-1-specific constantsbrian m. carlson, Feb 22, 2020
  9. Junio C HamanoFeb 24, 2020
  10. Jeff KingFeb 24, 2020
  11. Junio C HamanoFeb 24, 2020
  12. 12/24 setup: allow check_repository_format to read repository formatbrian m. carlson, Feb 22, 2020
  13. 11/24 t/helper: make repository tests hash independentbrian m. carlson, Feb 22, 2020
  14. 14/24 builtin/init-db: add environment variable for new repo hashbrian m. carlson, Feb 22, 2020
  15. 06/24 t: use hash-specific lookup tables to define test constantsbrian m. carlson, Feb 22, 2020
  16. 10/24 t/helper: initialize repository if necessarybrian m. carlson, Feb 22, 2020
  17. Junio C HamanoFeb 24, 2020
  18. brian m. carlsonFeb 25, 2020
  19. 22/24 fast-import: make find_marks work on any mark setbrian m. carlson, Feb 22, 2020
  20. 19/24 tag: store SHA-256 signatures in a headerbrian m. carlson, Feb 22, 2020
  21. Junio C HamanoFeb 24, 2020
  22. 23/24 fast-import: add a generic function to iterate over marksbrian m. carlson, Feb 22, 2020
  23. 13/24 builtin/init-db: allow specifying hash algorithm on command linebrian m. carlson, Feb 22, 2020
  24. Junio C HamanoFeb 24, 2020
  25. brian m. carlsonFeb 25, 2020
  26. 17/24 commit: use expected signature header for SHA-256brian m. carlson, Feb 22, 2020
  27. Junio C HamanoFeb 24, 2020
  28. 18/24 gpg-interface: improve interface for parsing tagsbrian m. carlson, Feb 22, 2020
  29. Junio C HamanoFeb 24, 2020
  30. Johannes SchindelinFeb 25, 2020
  31. Junio C HamanoFeb 25, 2020
  32. brian m. carlsonFeb 26, 2020
  33. Junio C HamanoFeb 26, 2020
  34. brian m. carlsonFeb 26, 2020
  35. Johannes SchindelinFeb 27, 2020
  36. Junio C HamanoFeb 27, 2020
  37. 20/24 fast-import: permit reading multiple marks filesbrian m. carlson, Feb 22, 2020
  38. Junio C HamanoJun 5, 2020
  39. Junio C HamanoJun 5, 2020
  40. brian m. carlsonJun 5, 2020
  41. Junio C HamanoJun 5, 2020
  42. 24/24 fast-import: add options for rewriting submodulesbrian m. carlson, Feb 22, 2020
  43. 21/24 fast-import: add helper function for inserting mark object entriesbrian m. carlson, Feb 22, 2020
  44. 04/24 hex: add functions to parse hex object IDs in any algorithmbrian m. carlson, Feb 22, 2020
  45. 16/24 worktree: allow repository version 1brian m. carlson, Feb 22, 2020
  46. 05/24 repository: require a build flag to use SHA-256brian m. carlson, Feb 22, 2020
  47. 02/24 hash: implement and use a context cloning functionbrian m. carlson, Feb 22, 2020
  48. 15/24 init-db: move writing repo version into a functionbrian m. carlson, Feb 22, 2020
  49. 08/24 t6300: make hash algorithm independentbrian m. carlson, Feb 22, 2020
  50. 09/24 t/helper/test-dump-split-index: initialize git repositorybrian m. carlson, Feb 22, 2020
  51. Junio C HamanoFeb 24, 2020

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.