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

Re: [PATCH 12/18] streaming: rely on object sources to create object stream

From
Karthik Nayak <karthik.188@gmail.com>
Date
Nov 19, 2025, 16:10 UTC
Message-ID
<CAOLa=ZRwnsYeHDpdL+uvnw0YMTbG1Gx2SKsq+0hTWMto+QZ+Lg@mail.gmail.com>
In-Reply-To
<20251119-b4-pks-odb-read-stream-v1-12-adacf03c2ccf@pks.im>
Patrick Steinhardt <ps@pks.im> writes:
Show 27 quoted lines
> When creating an object stream we first look up the object info and, if
> it's present, we call into the respective backend that contains the
> object to create a new stream for it.
>
> This has the consequence that, for loose object source, we basically
> iterate through the object sources twice: we first discover that the
> file exists as a loose object in the first place by iterating through
> all sources. And, once we have discovered it, we again walk through all
> sources to try and map the object. The same issue will eventually also
> surface once the packfile store becomes per-object-source.
>
> Furthermore, it feels rather pointless to first look up the object only
> to then try and read it.
>
> Refactor the logic to be centered around sources instead. Instead of
> first reading the object, we immediately ask the source to create the
> object stream for us. If the object exists we get stream, otherwise
> we'll try the next source.
>
> Like this we only have to iterate through sources once. But even more
> importantly, this change also helps us to make the whole logic
> pluggable. The object read stream subsystem does not need to be aware of
> the different source backends anymore, but eventually it'll only have to
> call the source's callback function.
>
> Note that at the current poin in time we aren't full there yet:
>

s/poin/point s/full/fully

Show 41 quoted lines
>   - The packfile store still sits on the object database level and is
>     thus agnostic of the sources.
>
>   - We still have to call into both the packfile store and the loose
>     object source.
>
> But both of these issues will soon be addressed.
>
> Signed-off-by: Patrick Steinhardt <ps@pks.im>
> ---
>  streaming.c | 65 +++++++++++++++++++++++--------------------------------------
>  1 file changed, 24 insertions(+), 41 deletions(-)
>
> diff --git a/streaming.c b/streaming.c
> index 572be98248..bebb434cd1 100644
> --- a/streaming.c
> +++ b/streaming.c
> @@ -204,21 +204,15 @@ static int close_istream_loose(struct odb_read_stream *_st)
>  }
>
>  static int open_istream_loose(struct odb_read_stream **out,
> -			      struct repository *r,
> +			      struct odb_source *source,
>  			      const struct object_id *oid)
>  {
>  	struct object_info oi = OBJECT_INFO_INIT;
>  	struct odb_loose_read_stream *st;
> -	struct odb_source *source;
>  	unsigned long mapsize;
>  	void *mapped;
>
> -	odb_prepare_alternates(r->objects);
> -	for (source = r->objects->sources; source; source = source->next) {
> -		mapped = odb_source_loose_map_object(source, oid, &mapsize);
> -		if (mapped)
> -			break;
> -	}
> +	mapped = odb_source_loose_map_object(source, oid, &mapsize);
>  	if (!mapped)
>  		return -1;
>

So instead of going over the sources, we simply check for the given source. Nice.

[snip]
Show 36 quoted lines
> @@ -462,30 +460,15 @@ static int istream_source(struct odb_read_stream **out,
>  			  struct repository *r,
>  			  const struct object_id *oid)
>  {
> -	unsigned long size;
> -	int status;
> -	struct object_info oi = OBJECT_INFO_INIT;
> -
> -	oi.sizep = &size;
> -	status = odb_read_object_info_extended(r->objects, oid, &oi, 0);
> -	if (status < 0)
> -		return status;
> +	struct odb_source *source;
>
> -	switch (oi.whence) {
> -	case OI_LOOSE:
> -		if (open_istream_loose(out, r, oid) < 0)
> -			break;
> -		return 0;
> -	case OI_PACKED:
> -		if (oi.u.packed.is_delta ||
> -		    repo_settings_get_big_file_threshold(the_repository) >= size ||
> -		    open_istream_pack_non_delta(out, r, oid, oi.u.packed.pack,
> -						oi.u.packed.offset) < 0)
> -			break;
> +	if (!open_istream_pack_non_delta(out, r->objects, oid))
>  		return 0;
> -	default:
> -		break;
> -	}
> +
> +	odb_prepare_alternates(r->objects);
> +	for (source = r->objects->sources; source; source = source->next)
> +		if (!open_istream_loose(out, source, oid))
> +			return 0;
>

This seem to be the crux of it, where earlier we depended on `odb_read_object_info_extended()` to tell us which backend to rely on and then we re-fetched from that backed, now we simply go over the different sources and try to get the object stream. Makes sense.

Show 5 quoted lines
>  	return open_istream_incore(out, r, oid);
>  }
>
> --
> 2.52.0.rc2.482.gaa765fefd0.dirty
Previous: Patrick SteinhardtNext: Patrick Steinhardt
Message 28 of 85 in “Refactor object read streams to work via object sources”
  1. 00/18 Refactor object read streams to work via object sourcesPatrick Steinhardt, Nov 19, 2025
  2. 01/18 streaming: rename `git_istream` into `odb_read_stream`Patrick Steinhardt, Nov 19, 2025
  3. Justin ToblerNov 19, 2025
  4. Junio C HamanoNov 19, 2025
  5. Patrick SteinhardtNov 21, 2025
  6. 02/18 streaming: drop the `open()` callback functionPatrick Steinhardt, Nov 19, 2025
  7. Karthik NayakNov 19, 2025
  8. Justin ToblerNov 19, 2025
  9. Patrick SteinhardtNov 21, 2025
  10. 03/18 streaming: propagate final object type via the streamPatrick Steinhardt, Nov 19, 2025
  11. Justin ToblerNov 19, 2025
  12. Patrick SteinhardtNov 21, 2025
  13. 04/18 streaming: explicitly pass packfile info when streaming a packed objectPatrick Steinhardt, Nov 19, 2025
  14. 05/18 streaming: allocate stream inside the backend-specific logicPatrick Steinhardt, Nov 19, 2025
  15. Karthik NayakNov 19, 2025
  16. Patrick SteinhardtNov 21, 2025
  17. 06/18 streaming: create structure for in-core object streamsPatrick Steinhardt, Nov 19, 2025
  18. Karthik NayakNov 19, 2025
  19. Patrick SteinhardtNov 21, 2025
  20. 07/18 streaming: create structure for loose object streamsPatrick Steinhardt, Nov 19, 2025
  21. 08/18 streaming: create structure for packed object streamsPatrick Steinhardt, Nov 19, 2025
  22. 09/18 streaming: create structure for filtered object streamsPatrick Steinhardt, Nov 19, 2025
  23. 10/18 streaming: move zlib stream into backendsPatrick Steinhardt, Nov 19, 2025
  24. 11/18 packfile: introduce function to read object info from a storePatrick Steinhardt, Nov 19, 2025
  25. Karthik NayakNov 19, 2025
  26. Patrick SteinhardtNov 21, 2025
  27. 12/18 streaming: rely on object sources to create object streamPatrick Steinhardt, Nov 19, 2025
  28. Karthik NayakNov 19, 2025
  29. 13/18 streaming: get rid of `the_repository`Patrick Steinhardt, Nov 19, 2025
  30. 14/18 streaming: make the `odb_read_stream` definition publicPatrick Steinhardt, Nov 19, 2025
  31. Karthik NayakNov 19, 2025
  32. Patrick SteinhardtNov 21, 2025
  33. 15/18 streaming: move logic to read loose objects streams into backendPatrick Steinhardt, Nov 19, 2025
  34. 16/18 streaming: move logic to read packed objects streams into backendPatrick Steinhardt, Nov 19, 2025
  35. 17/18 streaming: refactor interface to be object-database-centricPatrick Steinhardt, Nov 19, 2025
  36. 18/18 streaming: move into object database subsystemPatrick Steinhardt, Nov 19, 2025
  37. 00/19 Refactor object read streams to work via object sourcesPatrick Steinhardt, Nov 21, 2025
  38. 01/19 streaming: rename `git_istream` into `odb_read_stream`Patrick Steinhardt, Nov 21, 2025
  39. 02/19 streaming: drop the `open()` callback functionPatrick Steinhardt, Nov 21, 2025
  40. Junio C HamanoNov 21, 2025
  41. Patrick SteinhardtNov 23, 2025
  42. 03/19 streaming: propagate final object type via the streamPatrick Steinhardt, Nov 21, 2025
  43. 04/19 streaming: explicitly pass packfile info when streaming a packed objectPatrick Steinhardt, Nov 21, 2025
  44. 05/19 streaming: allocate stream inside the backend-specific logicPatrick Steinhardt, Nov 21, 2025
  45. 06/19 streaming: create structure for in-core object streamsPatrick Steinhardt, Nov 21, 2025
  46. 07/19 streaming: create structure for loose object streamsPatrick Steinhardt, Nov 21, 2025
  47. 08/19 streaming: create structure for packed object streamsPatrick Steinhardt, Nov 21, 2025
  48. 09/19 streaming: create structure for filtered object streamsPatrick Steinhardt, Nov 21, 2025
  49. 10/19 streaming: move zlib stream into backendsPatrick Steinhardt, Nov 21, 2025
  50. 11/19 packfile: introduce function to read object info from a storePatrick Steinhardt, Nov 21, 2025
  51. 12/19 streaming: rely on object sources to create object streamPatrick Steinhardt, Nov 21, 2025
  52. Junio C HamanoNov 21, 2025
  53. Patrick SteinhardtNov 23, 2025
  54. 13/19 streaming: get rid of `the_repository`Patrick Steinhardt, Nov 21, 2025
  55. Junio C HamanoNov 21, 2025
  56. Patrick SteinhardtNov 23, 2025
  57. 14/19 streaming: make the `odb_read_stream` definition publicPatrick Steinhardt, Nov 21, 2025
  58. 15/19 streaming: move logic to read loose objects streams into backendPatrick Steinhardt, Nov 21, 2025
  59. 16/19 streaming: move logic to read packed objects streams into backendPatrick Steinhardt, Nov 21, 2025
  60. 17/19 streaming: refactor interface to be object-database-centricPatrick Steinhardt, Nov 21, 2025
  61. Junio C HamanoNov 22, 2025
  62. Patrick SteinhardtNov 23, 2025
  63. 18/19 streaming: move into object database subsystemPatrick Steinhardt, Nov 21, 2025
  64. Junio C HamanoNov 23, 2025
  65. 19/19 streaming: drop redundant type and size pointersPatrick Steinhardt, Nov 21, 2025
  66. 00/19 Refactor object read streams to work via object sourcesPatrick Steinhardt, Nov 23, 2025
  67. 01/19 streaming: rename `git_istream` into `odb_read_stream`Patrick Steinhardt, Nov 23, 2025
  68. 02/19 streaming: drop the `open()` callback functionPatrick Steinhardt, Nov 23, 2025
  69. 03/19 streaming: propagate final object type via the streamPatrick Steinhardt, Nov 23, 2025
  70. 04/19 streaming: explicitly pass packfile info when streaming a packed objectPatrick Steinhardt, Nov 23, 2025
  71. 05/19 streaming: allocate stream inside the backend-specific logicPatrick Steinhardt, Nov 23, 2025
  72. 06/19 streaming: create structure for in-core object streamsPatrick Steinhardt, Nov 23, 2025
  73. 07/19 streaming: create structure for loose object streamsPatrick Steinhardt, Nov 23, 2025
  74. 08/19 streaming: create structure for packed object streamsPatrick Steinhardt, Nov 23, 2025
  75. 09/19 streaming: create structure for filtered object streamsPatrick Steinhardt, Nov 23, 2025
  76. 10/19 streaming: move zlib stream into backendsPatrick Steinhardt, Nov 23, 2025
  77. 11/19 packfile: introduce function to read object info from a storePatrick Steinhardt, Nov 23, 2025
  78. 12/19 streaming: rely on object sources to create object streamPatrick Steinhardt, Nov 23, 2025
  79. 13/19 streaming: get rid of `the_repository`Patrick Steinhardt, Nov 23, 2025
  80. 14/19 streaming: make the `odb_read_stream` definition publicPatrick Steinhardt, Nov 23, 2025
  81. 15/19 streaming: move logic to read loose objects streams into backendPatrick Steinhardt, Nov 23, 2025
  82. 16/19 streaming: move logic to read packed objects streams into backendPatrick Steinhardt, Nov 23, 2025
  83. 17/19 streaming: refactor interface to be object-database-centricPatrick Steinhardt, Nov 23, 2025
  84. 18/19 streaming: move into object database subsystemPatrick Steinhardt, Nov 23, 2025
  85. 19/19 streaming: drop redundant type and size pointersPatrick Steinhardt, Nov 23, 2025

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.