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

Re: [PATCH 05/10] setup: use the default algorithm to initialize repo format

From
Junio C Hamano <gitster@pobox.com>
Date
Jun 20, 2025, 14:55 UTC
Message-ID
<xmqqtt4a5upb.fsf@gitster.g>
In-Reply-To
<20250620011943.586596-6-sandals@crustytoothpaste.net>
"brian m. carlson" <sandals@crustytoothpaste.net> writes:
Show 8 quoted lines
> When we define a new repository format with REPOSITORY_FORMAT_INIT, we
> always use GIT_HASH_SHA1, and this value ends up getting used as the
> default value to initialize a repository if none of the command line,
> environment, or config tell us to do otherwise.
>
> Because we might not always want to use SHA-1 as the default, let's
> instead specify the default hash algorithm constant so that we will use
> whatever the specified default is.  
All of the above hints the use of _DEFAULT instead of _SHA1 but ...
> However, to make sure we continue to
> read repositories without a specified hash algorithm as SHA-1, default
> the repository format to the original hash algorithm (SHA-1) when
> reading the repository format.
... this explains why we may want to
 - expect that we would be able to read the hash from repository
 - read from repository
 - if the repository specifies the hash, happily use it
 - otherwise it is a lagacy repository so use the SHA-1

Is that what is going on here? Because I find some things that happens in the code somewhat questionable.

Show 15 quoted lines
> Signed-off-by: brian m. carlson <sandals@crustytoothpaste.net>
> ---
>  setup.c | 5 ++++-
>  setup.h | 2 +-
>  2 files changed, 5 insertions(+), 2 deletions(-)
>
> diff --git a/setup.c b/setup.c
> index 641c857ed5..fb38824a2b 100644
> --- a/setup.c
> +++ b/setup.c
> @@ -835,9 +835,12 @@ static void init_repository_format(struct repository_format *format)
>  int read_repository_format(struct repository_format *format, const char *path)
>  {
>  	clear_repository_format(format);
> +	format->hash_algo = GIT_HASH_ORIGINAL;

If we expect we can read from the config, and otherwise we fall back to the hardcoded legacy SHA-1, do we need this assignment? We cleared and version is set to -1, and then we read from the config ...

>  	git_config_from_file(check_repo_format, path, format);

... so if the file said anything about "extensions.objectformat", we would know it by now. If not, wouldn't version be left to -1 as our previous clal to clear_repository_format() set it via its call to init_repository_format()?

Ahh, this code is prepared to handle a repository that claims to use version 1 but does not set extensions.objectformat at all. And in such a case, we do want to use SHA-1. OK, the above code makes sense for that case.

> -	if (format->version == -1)
> +	if (format->version == -1) {

And if there is no core.repositoryformatversion set, we will come here. According to the comment before handle_extension_v0(), some extensions.* should still be honored even in such a repository, and the above call to git_config_from_file() should have handled them just fine.

However, I do not understand why we clear all of what we read with another call to clear_repository_format() here.

Show 5 quoted lines
>  		clear_repository_format(format);
> +		format->hash_algo = GIT_HASH_ORIGINAL;
> +	}
>  	return format->version;
>  }

Admittedly, I find that the way how check_repo_format() does its thing somewhat questionable. Even though it reads into the .version member the value of core.repositoryformatversion, it slurps in all the extensions.* regardless of what .version the repository claims to be in. So even though there are two separate functions to handle "historical compatibility" handle_extension_v0() and other extensions, we still end up honoring extensions.objectformat in a repository that does not say what format version it uses. And clearing them with a call to clear_repository_format() may make sense, but do we want to clear things we read with handle_extension_v0() as well?

Show 13 quoted lines
> diff --git a/setup.h b/setup.h
> index 18dc3b7368..8522fa8575 100644
> --- a/setup.h
> +++ b/setup.h
> @@ -149,7 +149,7 @@ struct repository_format {
>  { \
>  	.version = -1, \
>  	.is_bare = -1, \
> -	.hash_algo = GIT_HASH_SHA1, \
> +	.hash_algo = GIT_HASH_DEFAULT, \
>  	.ref_storage_format = REF_STORAGE_FORMAT_FILES, \
>  	.unknown_extensions = STRING_LIST_INIT_DUP, \
>  	.v1_only_extensions = STRING_LIST_INIT_DUP, \
Previous: brian m. carlsonNext: brian m. carlson
Message 18 of 47 in “Add SHA-256 by default as a breaking change”
  1. 00/10 Add SHA-256 by default as a breaking changebrian m. carlson, Jun 20, 2025
  2. 01/10 hash: add a constant for the default hash algorithmbrian m. carlson, Jun 20, 2025
  3. 03/10 builtin: use default hash when outside a repositorybrian m. carlson, Jun 20, 2025
  4. Junio C HamanoJun 20, 2025
  5. Patrick SteinhardtJul 1, 2025
  6. brian m. carlsonJul 1, 2025
  7. Patrick SteinhardtJul 2, 2025
  8. 04/10 Use original hash for legacy formatsbrian m. carlson, Jun 20, 2025
  9. Junio C HamanoJun 20, 2025
  10. brian m. carlsonJun 20, 2025
  11. Junio C HamanoJun 20, 2025
  12. Patrick SteinhardtJul 1, 2025
  13. 02/10 hash: add a constant for the original hash algorithmbrian m. carlson, Jun 20, 2025
  14. Junio C HamanoJun 20, 2025
  15. brian m. carlsonJun 20, 2025
  16. Patrick SteinhardtJul 1, 2025
  17. 05/10 setup: use the default algorithm to initialize repo formatbrian m. carlson, Jun 20, 2025
  18. Junio C HamanoJun 20, 2025
  19. brian m. carlsonJun 20, 2025
  20. Junio C HamanoJun 20, 2025
  21. 08/10 t4042: choose the built-in hash outside of a repobrian m. carlson, Jun 20, 2025
  22. 07/10 t1007: choose the built-in hash outside of a repobrian m. carlson, Jun 20, 2025
  23. 06/10 t: default to compile-time default hash if not setbrian m. carlson, Jun 20, 2025
  24. 09/10 t5300: choose the built-in hash outside of a repobrian m. carlson, Jun 20, 2025
  25. 10/10 Enable SHA-256 by default in breaking changes modebrian m. carlson, Jun 20, 2025
  26. Junio C HamanoJun 20, 2025
  27. brian m. carlsonJun 20, 2025
  28. Junio C HamanoJun 20, 2025
  29. brian m. carlsonJun 20, 2025
  30. Junio C HamanoJun 20, 2025
  31. brian m. carlsonJun 20, 2025
  32. Patrick SteinhardtJul 1, 2025
  33. 00/11 Add SHA-256 by default as a breaking changebrian m. carlson, Jul 1, 2025
  34. 02/11 hash: add a constant for the legacy hash algorithmbrian m. carlson, Jul 1, 2025
  35. 01/11 hash: add a constant for the default hash algorithmbrian m. carlson, Jul 1, 2025
  36. 04/11 Use legacy hash for legacy formatsbrian m. carlson, Jul 1, 2025
  37. 03/11 builtin: use default hash when outside a repositorybrian m. carlson, Jul 1, 2025
  38. 06/11 t: default to compile-time default hash if not setbrian m. carlson, Jul 1, 2025
  39. 05/11 setup: use the default algorithm to initialize repo formatbrian m. carlson, Jul 1, 2025
  40. 08/11 t4042: choose the built-in hash outside of a repobrian m. carlson, Jul 1, 2025
  41. 07/11 t1007: choose the built-in hash outside of a repobrian m. carlson, Jul 1, 2025
  42. 10/11 help: add a build option for default hashbrian m. carlson, Jul 1, 2025
  43. 11/11 Enable SHA-256 by default in breaking changes modebrian m. carlson, Jul 1, 2025
  44. 09/11 t5300: choose the built-in hash outside of a repobrian m. carlson, Jul 1, 2025
  45. Junio C HamanoJul 1, 2025
  46. Patrick SteinhardtJul 2, 2025
  47. Kristoffer HaugsbakkJul 2, 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.