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

Re: [PATCH 03/17] odb: embed base source in the "files" backend

From
Justin Tobler <jltobler@gmail.com>
Date
Mar 4, 2026, 17:40 UTC
Message-ID
<aahkh1ICViKjP6Il@denethor>
In-Reply-To
<20260223-b4-pks-odb-source-pluggable-v1-3-253bac1db598@pks.im>
On 26/02/23 05:17PM, Patrick Steinhardt wrote:
Show 19 quoted lines
> The "files" backend is implemented as a pointer in the `struct
> odb_source`. This contradicts our typical pattern for pluggable backends
> like we use it for example in the ref store or for object database
> streams, where we typically embed the generic base structure in the
> specialized implementation. This pattern has a couple of small benefits:
> 
>   - We avoid an extra allocation.
> 
>   - We hide implementation details in the generic structure.
> 
>   - We can easily downcast from a generic backend to the specialized
>     structure and vice versa because the offsets are known at compile
>     time.
> 
>   - It becomes trivial to identify locations where we depend on backend
>     specific logic because the cast needs to be explicit.
> 
> Refactor our "files" object database source to do the same and embed the
> `struct odb_source` in the `struct odb_source_files`.
Makes sense.
 
Show 13 quoted lines
> There are still a bunch of sites in our code base where we do have to
> access internals of the "files" backend. The intent is that those will
> go away over time, but this will certainly take a while. Meanwhile,
> provide a `odb_source_files_downcast()` function that can convert a
> generic source into a "files" source.
> 
> As we only have a single source the downcast succeeds unconditionally
> for now. Eventually though the intent is to make the cast `BUG()` in
> case the caller requests to downcast a non-"files" backend to a "files"
> backend.
> 
> Signed-off-by: Patrick Steinhardt <ps@pks.im>
> ---
[snip]
Show 33 quoted lines
> diff --git a/odb/source-files.c b/odb/source-files.c
> index cbdaa6850f..a43a197157 100644
> --- a/odb/source-files.c
> +++ b/odb/source-files.c
> @@ -1,5 +1,6 @@
>  #include "git-compat-util.h"
>  #include "object-file.h"
> +#include "odb/source.h"
>  #include "odb/source-files.h"
>  #include "packfile.h"
>  
> @@ -9,15 +10,20 @@ void odb_source_files_free(struct odb_source_files *files)
>  		return;
>  	odb_source_loose_free(files->loose);
>  	packfile_store_free(files->packed);
> +	odb_source_release(&files->base);
>  	free(files);
>  }
>  
> -struct odb_source_files *odb_source_files_new(struct odb_source *source)
> +struct odb_source_files *odb_source_files_new(struct object_database *odb,
> +					      const char *path,
> +					      bool local)
>  {
>  	struct odb_source_files *files;
> +
>  	CALLOC_ARRAY(files, 1);
> -	files->source = source;
> -	files->loose = odb_source_loose_new(source);
> -	files->packed = packfile_store_new(source);
> +	odb_source_init(&files->base, odb, path, local);
> +	files->loose = odb_source_loose_new(&files->base);
> +	files->packed = packfile_store_new(&files->base);

When creating the files ODB source, it is now responsible for also creating the embedded base ODB souce. Makes sense.

Show 24 quoted lines
> +
>  	return files;
>  }
> diff --git a/odb/source-files.h b/odb/source-files.h
> index 0b8bf773ca..58753d40de 100644
> --- a/odb/source-files.h
> +++ b/odb/source-files.h
> @@ -1,8 +1,9 @@
>  #ifndef ODB_SOURCE_FILES_H
>  #define ODB_SOURCE_FILES_H
>  
> +#include "odb/source.h"
> +
>  struct odb_source_loose;
> -struct odb_source;
>  struct packfile_store;
>  
>  /*
> @@ -10,15 +11,26 @@ struct packfile_store;
>   * packfiles. It is the default backend used by Git to store objects.
>   */
>  struct odb_source_files {
> -	struct odb_source *source;
> +	struct odb_source base;

Out of curiousity, was there any reason to the reference ODB source in the prior patch? Seems like we could have just added it here.

Show 17 quoted lines
>  	struct odb_source_loose *loose;
>  	struct packfile_store *packed;
>  };
>  
>  /* Allocate and initialize a new object source. */
> -struct odb_source_files *odb_source_files_new(struct odb_source *source);
> +struct odb_source_files *odb_source_files_new(struct object_database *odb,
> +					      const char *path,
> +					      bool local);
>  
>  /* Free the object source and release all associated resources. */
>  void odb_source_files_free(struct odb_source_files *files);
>  
> +/*
> + * Cast the given object database source to the files backend. This will cause
> + * a BUG in case the source doesn't use this backend.
> + */

In the commit message you mention that eventually `odb_source_files_downcast()` will BUG() if the source doesn't use the backend. But, it doesn't appear to do this yet. Should we still have this comment?

Show 6 quoted lines
> +static inline struct odb_source_files *odb_source_files_downcast(struct odb_source *source)
> +{
> +	return container_of(source, struct odb_source_files, base);
> +}
> +
>  #endif
[snip]
Show 19 quoted lines
> diff --git a/odb/source.h b/odb/source.h
> index 1c34265189..e6698b73a3 100644
> --- a/odb/source.h
> +++ b/odb/source.h
> @@ -1,8 +1,6 @@
>  #ifndef ODB_SOURCE_H
>  #define ODB_SOURCE_H
>  
> -#include "odb/source-files.h"
> -
>  /*
>   * The source is the part of the object database that stores the actual
>   * objects. It thus encapsulates the logic to read and write the specific
> @@ -21,9 +19,6 @@ struct odb_source {
>  	/* Object database that owns this object source. */
>  	struct object_database *odb;
>  
> -	/* The backend used to store objects. */
> -	struct odb_source_files *files;

Now that the base ODB source is embedded in `struct odb_source_files`, it is accessed via downcasting and the direct reference is no longer needed. This is responsible for most of the structural change fallout in this patch.

Show 35 quoted lines
> -
>  	/*
>  	 * Figure out whether this is the local source of the owning
>  	 * repository, which would typically be its ".git/objects" directory.
> @@ -53,7 +48,31 @@ struct odb_source *odb_source_new(struct object_database *odb,
>  				  const char *path,
>  				  bool local);
>  
> -/* Free the object database source, releasing all associated resources. */
> +/*
> + * Initialize the source for the given object database located at `path`.
> + * `local` indicates whether or not the source is the local and thus primary
> + * object source of the object database.
> + *
> + * This function is only supposed to be called by specific object source
> + * implementations.
> + */
> +void odb_source_init(struct odb_source *source,
> +		     struct object_database *odb,
> +		     const char *path,
> +		     bool local);
> +
> +/*
> + * Free the object database source, releasing all associated resources and
> + * freeing the structure itself.
> + */
>  void odb_source_free(struct odb_source *source);
>  
> +/*
> + * Release the object database source, releasing all associated resources.
> + *
> + * This function is only supposed to be called by specific object source
> + * implementations.
> + */
> +void odb_source_release(struct odb_source *source);

From a naming perspective, I do find the odb_source_new() vs odb_source_init() and odb_source_free() vs odb_source_release() interfaces to be tad bit confusing. I understand that odb_source_init() and odb_source_release() and only intended for use by the concrete ODB source implementations to facilitate initializing/freeing the base ODB source. The comments also do help clarify this, but I think it is still rather easy to get them mixed up when reading.

Maybe we could rename them to odb_base_source_init() and odb_base_source_free()?

Show 16 quoted lines
> +
>  #endif
> diff --git a/odb/streaming.c b/odb/streaming.c
> index 26b0a1a0f5..19cda9407d 100644
> --- a/odb/streaming.c
> +++ b/odb/streaming.c
> @@ -187,7 +187,8 @@ static int istream_source(struct odb_read_stream **out,
>  
>  	odb_prepare_alternates(odb);
>  	for (source = odb->sources; source; source = source->next) {
> -		if (!packfile_store_read_object_stream(out, source->files->packed, oid) ||
> +		struct odb_source_files *files = odb_source_files_downcast(source);
> +		if (!packfile_store_read_object_stream(out, files->packed, oid) ||
>  		    !odb_source_loose_read_object_stream(out, source, oid))
>  			return 0;
>  	}
Overall this patch looks good.
-Justin
Previous: Justin ToblerNext: Justin Tobler
Message 24 of 77 in “odb: make object database sources pluggable”
  1. 00/17 odb: make object database sources pluggablePatrick Steinhardt, Feb 23, 2026
  2. 01/17 odb: split `struct odb_source` into separate headerPatrick Steinhardt, Feb 23, 2026
  3. 02/17 odb: introduce "files" sourcePatrick Steinhardt, Feb 23, 2026
  4. 03/17 odb: embed base source in the "files" backendPatrick Steinhardt, Feb 23, 2026
  5. 04/17 odb: move reparenting logic into respective subsystemsPatrick Steinhardt, Feb 23, 2026
  6. 05/17 odb/source: introduce source type for robustnessPatrick Steinhardt, Feb 23, 2026
  7. 06/17 odb/source: make `free()` function pluggablePatrick Steinhardt, Feb 23, 2026
  8. 07/17 odb/source: make `reprepare()` function pluggablePatrick Steinhardt, Feb 23, 2026
  9. 08/17 odb/source: make `close()` function pluggablePatrick Steinhardt, Feb 23, 2026
  10. 09/17 odb/source: make `read_object_info()` function pluggablePatrick Steinhardt, Feb 23, 2026
  11. 10/17 odb/source: make `read_object_stream()` function pluggablePatrick Steinhardt, Feb 23, 2026
  12. 11/17 odb/source: make `for_each_object()` function pluggablePatrick Steinhardt, Feb 23, 2026
  13. 12/17 odb/source: make `freshen_object()` function pluggablePatrick Steinhardt, Feb 23, 2026
  14. 13/17 odb/source: make `write_object()` function pluggablePatrick Steinhardt, Feb 23, 2026
  15. 14/17 odb/source: make `write_object_stream()` function pluggablePatrick Steinhardt, Feb 23, 2026
  16. 15/17 odb/source: make `read_alternates()` function pluggablePatrick Steinhardt, Feb 23, 2026
  17. 16/17 odb/source: make `write_alternate()` function pluggablePatrick Steinhardt, Feb 23, 2026
  18. 17/17 odb/source: make `begin_transaction()` function pluggablePatrick Steinhardt, Feb 23, 2026
  19. Patrick SteinhardtFeb 23, 2026
  20. Junio C HamanoFeb 23, 2026
  21. Patrick SteinhardtFeb 24, 2026
  22. Justin ToblerMar 4, 2026
  23. Justin ToblerMar 4, 2026
  24. Justin ToblerMar 4, 2026
  25. Justin ToblerMar 4, 2026
  26. Justin ToblerMar 4, 2026
  27. Justin ToblerMar 4, 2026
  28. Justin ToblerMar 4, 2026
  29. Justin ToblerMar 4, 2026
  30. Justin ToblerMar 4, 2026
  31. Justin ToblerMar 4, 2026
  32. Justin ToblerMar 4, 2026
  33. Karthik NayakMar 5, 2026
  34. Karthik NayakMar 5, 2026
  35. Karthik NayakMar 5, 2026
  36. Karthik NayakMar 5, 2026
  37. Karthik NayakMar 5, 2026
  38. Karthik NayakMar 5, 2026
  39. Karthik NayakMar 5, 2026
  40. Patrick SteinhardtMar 5, 2026
  41. Karthik NayakMar 5, 2026
  42. Patrick SteinhardtMar 5, 2026
  43. Patrick SteinhardtMar 5, 2026
  44. Patrick SteinhardtMar 5, 2026
  45. Patrick SteinhardtMar 5, 2026
  46. Patrick SteinhardtMar 5, 2026
  47. Patrick SteinhardtMar 5, 2026
  48. Patrick SteinhardtMar 5, 2026
  49. Patrick SteinhardtMar 5, 2026
  50. Patrick SteinhardtMar 5, 2026
  51. Patrick SteinhardtMar 5, 2026
  52. Patrick SteinhardtMar 5, 2026
  53. Patrick SteinhardtMar 5, 2026
  54. 00/17 odb: make object database sources pluggablePatrick Steinhardt, Mar 5, 2026
  55. 01/17 odb: split `struct odb_source` into separate headerPatrick Steinhardt, Mar 5, 2026
  56. 02/17 odb: introduce "files" sourcePatrick Steinhardt, Mar 5, 2026
  57. 03/17 odb: embed base source in the "files" backendPatrick Steinhardt, Mar 5, 2026
  58. 04/17 odb: move reparenting logic into respective subsystemsPatrick Steinhardt, Mar 5, 2026
  59. 05/17 odb/source: introduce source type for robustnessPatrick Steinhardt, Mar 5, 2026
  60. 06/17 odb/source: make `free()` function pluggablePatrick Steinhardt, Mar 5, 2026
  61. 07/17 odb/source: make `reprepare()` function pluggablePatrick Steinhardt, Mar 5, 2026
  62. 08/17 odb/source: make `close()` function pluggablePatrick Steinhardt, Mar 5, 2026
  63. 09/17 odb/source: make `read_object_info()` function pluggablePatrick Steinhardt, Mar 5, 2026
  64. 10/17 odb/source: make `read_object_stream()` function pluggablePatrick Steinhardt, Mar 5, 2026
  65. 11/17 odb/source: make `for_each_object()` function pluggablePatrick Steinhardt, Mar 5, 2026
  66. 12/17 odb/source: make `freshen_object()` function pluggablePatrick Steinhardt, Mar 5, 2026
  67. 13/17 odb/source: make `write_object()` function pluggablePatrick Steinhardt, Mar 5, 2026
  68. 14/17 odb/source: make `write_object_stream()` function pluggablePatrick Steinhardt, Mar 5, 2026
  69. 15/17 odb/source: make `read_alternates()` function pluggablePatrick Steinhardt, Mar 5, 2026
  70. 16/17 odb/source: make `write_alternate()` function pluggablePatrick Steinhardt, Mar 5, 2026
  71. 17/17 odb/source: make `begin_transaction()` function pluggablePatrick Steinhardt, Mar 5, 2026
  72. Justin ToblerMar 5, 2026
  73. Justin ToblerMar 5, 2026
  74. Justin ToblerMar 5, 2026
  75. Justin ToblerMar 5, 2026
  76. Junio C HamanoMar 5, 2026
  77. Patrick SteinhardtMar 10, 2026

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.