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

Re: [PATCH v11 26/41] walker.c: use ref transaction for ref updates

From
Jonathan Nieder <jrnieder@gmail.com>
Date
May 28, 2014, 19:56 UTC
Message-ID
<20140528195645.GY12314@google.com>
In-Reply-To
<1401222360-21175-27-git-send-email-sahlberg@google.com>
Ronnie Sahlberg wrote:
> Switch to using ref transactions in walker_fetch(). As part of the refactoring
> to use ref transactions we also fix a potential memory leak where in the
> original code if write_ref_sha1() would fail we would end up returning from
> the function without free()ing the msg string.
Sounds good.
> This changes the locking slightly for walker_fetch. Previously the code would
> lock all refs before writing them but now we do not lock the refs until the
> commit stage.

I don't think this actually changes locking in any significant way. (Timing changes, but that's to be expected whenever code changes.)

The old contract was:

You run git http-fetch with refnames passed with "-w". http-fetch will overwrite the refs with their new values.

The new contract is:

You run git http-fetch with refnames passed with "-w". http-fetch will overwrite the refs with their new values.

What changed?
[...]
> Note that this function is only called when fetching from a remote HTTP
> repository onto the local (most of the time single-user) repository which
> likely means that the type of collissions that the previous locking would
> protect against and cause the fetch to fail for to be even more rare.

More to the point, while this function is used by 'git remote-http' (and hence by "git fetch https://foo/bar/baz"), it is only used to fetch objects by that code path. The remote helper machinery handles the ref updates on its own (and passes NULL as the write_ref argument to the dumb walker).

[...]
> Signed-off-by: Ronnie Sahlberg <sahlberg@google.com>
> ---
>  walker.c | 56 +++++++++++++++++++++++++++++++++-----------------------
>  1 file changed, 33 insertions(+), 23 deletions(-)
The code change looks good from a quick glance.

gcc 4.6.3 doesn't seem to be smart enough to notice that 'transaction' is always initialized before it is used, so it (optionally) could be worth initializing 'transition = NULL' to work around that.

So I think this just needs a simpler commit message and then it would be good to go.

Jonathan
Previous: Ronnie SahlbergNext: Ronnie Sahlberg
Message 59 of 89 in “Use ref transactions”
  1. 00/41 Use ref transactionsRonnie Sahlberg, May 27, 2014
  2. 01/41 refs.c: remove ref_transaction_rollbackRonnie Sahlberg, May 27, 2014
  3. 02/41 refs.c: ref_transaction_commit should not free the transactionRonnie Sahlberg, May 27, 2014
  4. Jonathan NiederMay 27, 2014
  5. 03/41 refs.c: constify the sha arguments for ref_transaction_create|delete|updateRonnie Sahlberg, May 27, 2014
  6. 04/41 refs.c: allow passing NULL to ref_transaction_freeRonnie Sahlberg, May 27, 2014
  7. 05/41 refs.c: add a strbuf argument to ref_transaction_commit for error loggingRonnie Sahlberg, May 27, 2014
  8. 06/41 refs.c: add an err argument to repack_without_refsRonnie Sahlberg, May 27, 2014
  9. Jonathan NiederMay 28, 2014
  10. Ronnie SahlbergMay 28, 2014
  11. Jonathan NiederMay 28, 2014
  12. 07/41 refs.c: make ref_update_reject_duplicates take a strbuf argument for errorsRonnie Sahlberg, May 27, 2014
  13. 08/41 refs.c: add an err argument to delete_ref_looseRonnie Sahlberg, May 27, 2014
  14. Jonathan NiederMay 28, 2014
  15. Ronnie SahlbergMay 28, 2014
  16. 09/41 refs.c: make update_ref_write update a strbuf on failureRonnie Sahlberg, May 27, 2014
  17. 10/41 update-ref.c: log transaction error from the update_refRonnie Sahlberg, May 27, 2014
  18. Jonathan NiederMay 28, 2014
  19. Ronnie SahlbergMay 28, 2014
  20. Jonathan NiederMay 28, 2014
  21. 11/41 refs.c: remove the onerr argument to ref_transaction_commitRonnie Sahlberg, May 27, 2014
  22. 12/41 refs.c: change ref_transaction_update() to do error checking and return statusRonnie Sahlberg, May 27, 2014
  23. 13/41 refs.c: change ref_transaction_create to do error checking and return statusRonnie Sahlberg, May 27, 2014
  24. Jonathan NiederMay 28, 2014
  25. Ronnie SahlbergMay 28, 2014
  26. Jonathan NiederMay 28, 2014
  27. Ronnie SahlbergMay 28, 2014
  28. Jonathan NiederMay 28, 2014
  29. Ronnie SahlbergMay 28, 2014
  30. Jonathan NiederMay 28, 2014
  31. 14/41 refs.c: update ref_transaction_delete to check for error and return statusRonnie Sahlberg, May 27, 2014
  32. Jonathan NiederMay 28, 2014
  33. 15/41 refs.c: make ref_transaction_begin take an err argumentRonnie Sahlberg, May 27, 2014
  34. 16/41 refs.c: add transaction.status and track OPEN/CLOSED/ERRORRonnie Sahlberg, May 27, 2014
  35. Jonathan NiederMay 28, 2014
  36. Ronnie SahlbergMay 28, 2014
  37. 17/41 tag.c: use ref transactions when doing updatesRonnie Sahlberg, May 27, 2014
  38. 18/41 replace.c: use the ref transaction functions for updatesRonnie Sahlberg, May 27, 2014
  39. 19/41 commit.c: use ref transactions for updatesRonnie Sahlberg, May 27, 2014
  40. 20/41 sequencer.c: use ref transactions for all ref updatesRonnie Sahlberg, May 27, 2014
  41. Jonathan NiederMay 28, 2014
  42. 21/41 fast-import.c: change update_branch to use ref transactionsRonnie Sahlberg, May 27, 2014
  43. Jonathan NiederMay 28, 2014
  44. 22/41 branch.c: use ref transaction for all ref updatesRonnie Sahlberg, May 27, 2014
  45. 23/41 refs.c: change update_ref to use a transactionRonnie Sahlberg, May 27, 2014
  46. Jonathan NiederMay 28, 2014
  47. Ronnie SahlbergMay 28, 2014
  48. 24/41 receive-pack.c: use a reference transaction for updating the refsRonnie Sahlberg, May 27, 2014
  49. 25/41 fast-import.c: use a ref transaction when dumping tagsRonnie Sahlberg, May 27, 2014
  50. Jonathan NiederMay 28, 2014
  51. Ronnie SahlbergMay 28, 2014
  52. Jonathan NiederMay 28, 2014
  53. Ronnie SahlbergMay 28, 2014
  54. Jonathan NiederMay 28, 2014
  55. Ronnie SahlbergMay 29, 2014
  56. Jonathan NiederMay 29, 2014
  57. Ronnie SahlbergJun 3, 2014
  58. 26/41 walker.c: use ref transaction for ref updatesRonnie Sahlberg, May 27, 2014
  59. Jonathan NiederMay 28, 2014
  60. Ronnie SahlbergMay 28, 2014
  61. 27/41 refs.c: make lock_ref_sha1 staticRonnie Sahlberg, May 27, 2014
  62. 28/41 refs.c: remove the update_ref_lock functionRonnie Sahlberg, May 27, 2014
  63. 29/41 refs.c: remove the update_ref_write functionRonnie Sahlberg, May 27, 2014
  64. Jonathan NiederMay 28, 2014
  65. 30/41 refs.c: remove lock_ref_sha1Ronnie Sahlberg, May 27, 2014
  66. 31/41 refs.c: make prune_ref use a transaction to delete the refRonnie Sahlberg, May 27, 2014
  67. Jonathan NiederMay 28, 2014
  68. Ronnie SahlbergMay 28, 2014
  69. 32/41 refs.c: make delete_ref use a transactionRonnie Sahlberg, May 27, 2014
  70. Jonathan NiederMay 30, 2014
  71. Ronnie SahlbergJun 3, 2014
  72. 33/41 refs.c: pass the ref log message to _create/delete/update instead of _commitRonnie Sahlberg, May 27, 2014
  73. Jonathan NiederMay 30, 2014
  74. Ronnie SahlbergJun 3, 2014
  75. 34/41 refs.c: pass NULL as *flags to read_ref_fullRonnie Sahlberg, May 27, 2014
  76. 35/41 refs.c: pack all refs before we start to rename a refRonnie Sahlberg, May 27, 2014
  77. Jonathan NiederMay 30, 2014
  78. Ronnie SahlbergJun 3, 2014
  79. 36/41 refs.c: move the check for valid refname to lock_ref_sha1_basicRonnie Sahlberg, May 27, 2014
  80. Jonathan NiederMay 30, 2014
  81. Ronnie SahlbergJun 3, 2014
  82. 37/41 refs.c: call lock_ref_sha1_basic directly from commitRonnie Sahlberg, May 27, 2014
  83. 38/41 refs.c: pass a skip list to name_conflict_fnRonnie Sahlberg, May 27, 2014
  84. 39/41 refs.c: propagate any errno==ENOTDIR from _commit back to the callersRonnie Sahlberg, May 27, 2014
  85. Jonathan NiederMay 31, 2014
  86. Ronnie SahlbergJun 3, 2014
  87. 40/41 fetch.c: change s_update_ref to use a ref transactionRonnie Sahlberg, May 27, 2014
  88. 41/41 refs.c: make write_ref_sha1 staticRonnie Sahlberg, May 27, 2014
  89. Jonathan NiederJun 3, 2014

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.