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

[PATCH v5 0/7] Improvements for reading object info

From
Patrick Steinhardt <ps@pks.im>
Date
Jan 12, 2026, 09:00 UTC
Message-ID
<20260112-b4-pks-odb-read-object-info-improvements-v5-0-9a6124e95bf2@pks.im>
In-Reply-To
<20251218-b4-pks-odb-read-object-info-improvements-v1-0-81c8368492be@pks.im>
Hi,

this patch series contains various small improvements for reading object info for either loose or packed objects. These improvements were split out of a larger patch series where I'm about to introduce a new generic `odb_for_each_object()` function.

Changes in v5:
  - I discovered that this patch series incidentally fixes a segfault
    when using git-archive(1) to read deltified blobs that are larger
    than "core.bigFileThreshold". So the only change is an added test
    case that will detect this regression going forward.
  - Link to v4: https://lore.kernel.org/r/20260107-b4-pks-odb-read-object-info-improvements-v4-0-b5d55c47082a@pks.im
Changes in v4:
  - Extend the fix for OI_LOOSE and refactor the whole function to have
    a single exit path as proposed by Karthik. This results in a lot
    more changes, but makes the function way easier to reason about
    going forward.
  - Link to v3: https://lore.kernel.org/r/20260106-b4-pks-odb-read-object-info-improvements-v3-0-b5e02fae1fb0@pks.im
Changes in v3:
  - Fix a commit message typo.
  - Fix a function comment missing some words.
  - Link to v2: https://lore.kernel.org/r/20251218-b4-pks-odb-read-object-info-improvements-v2-0-62e3e49072bc@pks.im
Changes in v2:
  - Rebase the series on top of master with jc/object-read-stream-fix
    merged into it. I've also evicted the patch that fixes the same
    underlying issue.
  - Improve the commit message that drops OI_DBCACHED to explain why
    this is a safe refactoring.
  - Link to v1: https://lore.kernel.org/r/20251218-b4-pks-odb-read-object-info-improvements-v1-0-81c8368492be@pks.im
Thanks!
Patrick
---
Patrick Steinhardt (7):
      object-file: always set OI_LOOSE when reading object info
      packfile: always declare object info to be OI_PACKED
      packfile: extend `is_delta` field to allow for "unknown" state
      packfile: always populate pack-specific info when reading object info
      packfile: disentangle return value of `packed_object_info()`
      packfile: skip unpacking object header for disk size requests
      packfile: drop repository parameter from `packed_object_info()`
 builtin/cat-file.c     |   3 +-
 builtin/pack-objects.c |   4 +-
 commit-graph.c         |   2 +-
 object-file.c          | 115 ++++++++++++++++++++++++++++++-------------------
 odb.h                  |   8 +++-
 pack-bitmap.c          |   3 +-
 packfile.c             |  61 +++++++++++++++-----------
 packfile.h             |   7 ++-
 t/t5003-archive-zip.sh |  34 +++++++++++++++
 9 files changed, 158 insertions(+), 79 deletions(-)
Range-diff versus v4:
1:  07f529a631 = 1:  da9d514001 object-file: always set OI_LOOSE when reading object info
2:  b547df2885 ! 2:  c7b29f3789 packfile: always declare object info to be OI_PACKED
    @@ Commit message
         Drop the OI_DBCACHED enum completely. None of the callers seem to care
         about the distinction.
     
    +    Note that this also fixes a segfault introduced in 8c1b84bc97
    +    (streaming: move logic to read packed objects streams into backend,
    +    2025-11-23), which refactors how we stream packed objects. The intent is
    +    to only read packed objects in case they are stored non-deltified as
    +    we'd otherwise have to deflate them first. But the check for whether or
    +    not the object is stored as a delta was unconditionally done via
    +    `oi.u.packed.is_delta`, which is only valid in case `oi.whence` is
    +    `OI_PACKED`. But under some circumstances we got `OI_DBCACHED` here,
    +    which means that none of the `oi.u.packed` fields were initialized at
    +    all. Consequently, we assumed the object was not stored as a delta, and
    +    then try to read the object from `oi.u.packed.pack`, which is a `NULL`
    +    pointer and thus causes a segfault.
    +
    +    Add a test case for this issue so that this cannot regress in the
    +    future anymore.
    +
    +    Reported-by: Matt Smiley <msmiley@gitlab.com>
         Signed-off-by: Patrick Steinhardt <ps@pks.im>
     
      ## odb.h ##
    @@ packfile.c: int packed_object_info(struct repository *r, struct packed_git *p,
      
      out:
      	unuse_pack(&w_curs);
    +
    + ## t/t5003-archive-zip.sh ##
    +@@ t/t5003-archive-zip.sh: check_zip with_untracked2
    + check_added with_untracked2 untracked one/untracked
    + check_added with_untracked2 untracked two/untracked
    + 
    ++test_expect_success 'git-archive --format=zip with bigFile delta chains' '
    ++	test_when_finished rm -rf repo &&
    ++	git init repo &&
    ++	(
    ++		cd repo &&
    ++		test-tool genrandom foo 100000 >base &&
    ++		{
    ++			cat base &&
    ++			echo "trailing data"
    ++		} >delta-1 &&
    ++		{
    ++			cat delta-1 &&
    ++			echo "trailing data"
    ++		} >delta-2 &&
    ++		git add . &&
    ++		git commit -m "blobs" &&
    ++		git repack -Ad &&
    ++		git verify-pack -v .git/objects/pack/pack-*.idx >stats &&
    ++		test_grep "chain length = 1: 1 object" stats &&
    ++		test_grep "chain length = 2: 1 object" stats &&
    ++
    ++		git -c core.bigFileThreshold=1k archive --format=zip HEAD >archive.zip &&
    ++		if test_have_prereq UNZIP
    ++		then
    ++			mkdir unpack &&
    ++			cd unpack &&
    ++			"$GIT_UNZIP" ../archive.zip &&
    ++			test_cmp base ../base &&
    ++			test_cmp delta-1 ../delta-1 &&
    ++			test_cmp delta-2 ../delta-2
    ++		fi
    ++	)
    ++'
    ++
    + # Test remote archive over HTTP protocol.
    + #
    + # Note: this should be the last part of this test suite, because
3:  28940ce932 = 3:  ef5ac585f0 packfile: extend `is_delta` field to allow for "unknown" state
4:  c13c74467d = 4:  2a844d61fe packfile: always populate pack-specific info when reading object info
5:  d3c17fcc71 = 5:  a23f59d530 packfile: disentangle return value of `packed_object_info()`
6:  1c598686c5 = 6:  f246dc3745 packfile: skip unpacking object header for disk size requests
7:  afc5d85991 = 7:  a0c4f59547 packfile: drop repository parameter from `packed_object_info()`

--- base-commit: 7df68b50e49b6a1b576abb19b2e5d457749bc28b change-id: 20251215-b4-pks-odb-read-object-info-improvements-0e031ef827d2

Previous: Karthik NayakNext: Patrick Steinhardt
Message 50 of 58 in “Improvements for reading object info”
  1. 0/8 Improvements for reading object infoPatrick Steinhardt, Dec 18, 2025
  2. 1/8 object-file: always set OI_LOOSE when reading object infoPatrick Steinhardt, Dec 18, 2025
  3. 2/8 packfile: always declare object info to be OI_PACKEDPatrick Steinhardt, Dec 18, 2025
  4. Junio C HamanoDec 18, 2025
  5. Patrick SteinhardtDec 18, 2025
  6. 3/8 packfile: extend `is_delta` field to allow for "unknown" statePatrick Steinhardt, Dec 18, 2025
  7. 4/8 packfile: always populate pack-specific info when reading object infoPatrick Steinhardt, Dec 18, 2025
  8. Junio C HamanoDec 18, 2025
  9. 5/8 packfile: disentangle return value of `packed_object_info()`Patrick Steinhardt, Dec 18, 2025
  10. 6/8 packfile: skip unpacking object header for disk size requestsPatrick Steinhardt, Dec 18, 2025
  11. 7/8 packfile: fix short-circuiting of empty requestsPatrick Steinhardt, Dec 18, 2025
  12. 8/8 packfile: drop repository parameter from `packed_object_info()`Patrick Steinhardt, Dec 18, 2025
  13. Junio C HamanoDec 18, 2025
  14. Patrick SteinhardtDec 18, 2025
  15. 0/7 Improvements for reading object infoPatrick Steinhardt, Dec 18, 2025
  16. 1/7 object-file: always set OI_LOOSE when reading object infoPatrick Steinhardt, Dec 18, 2025
  17. 2/7 packfile: always declare object info to be OI_PACKEDPatrick Steinhardt, Dec 18, 2025
  18. Toon ClaesJan 5, 2026
  19. 3/7 packfile: extend `is_delta` field to allow for "unknown" statePatrick Steinhardt, Dec 18, 2025
  20. Toon ClaesJan 5, 2026
  21. Patrick SteinhardtJan 6, 2026
  22. 4/7 packfile: always populate pack-specific info when reading object infoPatrick Steinhardt, Dec 18, 2025
  23. Kristoffer HaugsbakkDec 30, 2025
  24. Patrick SteinhardtJan 5, 2026
  25. 5/7 packfile: disentangle return value of `packed_object_info()`Patrick Steinhardt, Dec 18, 2025
  26. 6/7 packfile: skip unpacking object header for disk size requestsPatrick Steinhardt, Dec 18, 2025
  27. 7/7 packfile: drop repository parameter from `packed_object_info()`Patrick Steinhardt, Dec 18, 2025
  28. 0/7 Improvements for reading object infoPatrick Steinhardt, Jan 6, 2026
  29. 1/7 object-file: always set OI_LOOSE when reading object infoPatrick Steinhardt, Jan 6, 2026
  30. Karthik NayakJan 7, 2026
  31. Patrick SteinhardtJan 7, 2026
  32. 2/7 packfile: always declare object info to be OI_PACKEDPatrick Steinhardt, Jan 6, 2026
  33. 3/7 packfile: extend `is_delta` field to allow for "unknown" statePatrick Steinhardt, Jan 6, 2026
  34. Karthik NayakJan 7, 2026
  35. 4/7 packfile: always populate pack-specific info when reading object infoPatrick Steinhardt, Jan 6, 2026
  36. 5/7 packfile: disentangle return value of `packed_object_info()`Patrick Steinhardt, Jan 6, 2026
  37. 6/7 packfile: skip unpacking object header for disk size requestsPatrick Steinhardt, Jan 6, 2026
  38. 7/7 packfile: drop repository parameter from `packed_object_info()`Patrick Steinhardt, Jan 6, 2026
  39. Karthik NayakJan 7, 2026
  40. 0/7 Improvements for reading object infoPatrick Steinhardt, Jan 7, 2026
  41. 1/7 object-file: always set OI_LOOSE when reading object infoPatrick Steinhardt, Jan 7, 2026
  42. Karthik NayakJan 8, 2026
  43. 2/7 packfile: always declare object info to be OI_PACKEDPatrick Steinhardt, Jan 7, 2026
  44. 3/7 packfile: extend `is_delta` field to allow for "unknown" statePatrick Steinhardt, Jan 7, 2026
  45. 4/7 packfile: always populate pack-specific info when reading object infoPatrick Steinhardt, Jan 7, 2026
  46. 5/7 packfile: disentangle return value of `packed_object_info()`Patrick Steinhardt, Jan 7, 2026
  47. 6/7 packfile: skip unpacking object header for disk size requestsPatrick Steinhardt, Jan 7, 2026
  48. 7/7 packfile: drop repository parameter from `packed_object_info()`Patrick Steinhardt, Jan 7, 2026
  49. Karthik NayakJan 8, 2026
  50. 0/7 Improvements for reading object infoPatrick Steinhardt, Jan 12, 2026
  51. 1/7 object-file: always set OI_LOOSE when reading object infoPatrick Steinhardt, Jan 12, 2026
  52. 2/7 packfile: always declare object info to be OI_PACKEDPatrick Steinhardt, Jan 12, 2026
  53. Junio C HamanoJan 12, 2026
  54. 3/7 packfile: extend `is_delta` field to allow for "unknown" statePatrick Steinhardt, Jan 12, 2026
  55. 4/7 packfile: always populate pack-specific info when reading object infoPatrick Steinhardt, Jan 12, 2026
  56. 5/7 packfile: disentangle return value of `packed_object_info()`Patrick Steinhardt, Jan 12, 2026
  57. 6/7 packfile: skip unpacking object header for disk size requestsPatrick Steinhardt, Jan 12, 2026
  58. 7/7 packfile: drop repository parameter from `packed_object_info()`Patrick Steinhardt, Jan 12, 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.