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

Re: Revision walking, commit dates, slop

From
Derrick Stolee <stolee@gmail.com>
Date
May 20, 2019, 11:20 UTC
Message-ID
<cfa2c367-5cd7-add5-0293-caa75b103f34@gmail.com>
In-Reply-To
<86mujhpewj.fsf@gmail.com>
On 5/20/2019 7:02 AM, Jakub Narebski wrote:
Show 10 quoted lines
>
> Are there any blockers that prevent the switch to this
> "generation number v2"?
> 
> - Is it a problem with insufficient data to choose the correct numbering
>   as "generation number v2' (there can be only one)?
> - Is it a problem with selected "generation number v2" being
>   incompatibile with gen v2, and Git failing when new version of
>   commit-graph is used instead of softly just not using commit-graph?
> - Or is it something else?
The switch becomes a bit complicated.

First, the plan was to version the commit-graph file to v2, and that would include a byte in the header for the "reachability index version" [1]. Since older clients fail hard on a newer file version, we are switching to instead including the reachability index version as a value in a more flexible "metadata chunk" [2]. Using the generation number column for the corrected commit-date offsets (assuming we also guarantee the offset is strictly increasing from parent to child), these new values will be backwards- compatible _except_ for 'git commit-graph verify'.

Second, we need to pull the reachability index value into a commit slab. The generation value is currently 32 bits, but we will expand that to 64 as it stores a timestamp. The commit struct is a bit bloated already, so this will reduce the required memory space even when not using the commit-graph. But, it requires some refactoring, including every place where we pass a "min_generation" needs to change type and name.

Third and finally, we need to calculate the new values and change the read logic to sum the offset and commit-date (when the metadata chunk says we are using corrected commit date).

While none of this is _incredibly_ hard to do, it does require a bit of care. It's on my list to get to at some point, but making the file incremental is higher priority to me.

Thanks, -Stolee

[1] https://public-inbox.org/git/pull.112.git.gitgitgadget@gmail.com/ [2] https://public-inbox.org/git/87h8acivkh.fsf@evledraar.gmail.com/

Previous: Jakub NarebskiNext: Jakub Narebski
Message 9 of 23 in “Revision walking, commit dates, slop”
  1. Mike HommeyMay 18, 2019
  2. SZEDER GáborMay 18, 2019
  3. Mike HommeyMay 18, 2019
  4. Mike HommeyMay 18, 2019
  5. SZEDER GáborMay 18, 2019
  6. Jakub NarebskiMay 19, 2019
  7. Derrick StoleeMay 20, 2019
  8. Jakub NarebskiMay 20, 2019
  9. Derrick StoleeMay 20, 2019
  10. Jakub NarebskiMay 20, 2019
  11. Jakub NarebskiMay 20, 2019
  12. Derrick StoleeMay 21, 2019
  13. Jakub NarebskiMay 22, 2019
  14. Derrick StoleeMay 22, 2019
  15. Jakub NarebskiMay 23, 2019
  16. Jakub NarebskiJun 25, 2019
  17. Derrick StoleeJun 25, 2019
  18. commit-graph: generation v5 (backward compatible date ceiling)Jakub Narebski, Sep 18, 2019
  19. Derrick StoleeSep 18, 2019
  20. revision: use generation for A..B --topo-order queriesDerrick Stolee, May 21, 2019
  21. 2/2 revision: keep topo-walk free of unintersting commitsDerrick Stolee, May 21, 2019
  22. Mike HommeyMay 22, 2019
  23. Jonathan NiederMay 21, 2019

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.