revset: Implement 'oldest' function that complements the 'latest' function in the revset language

[AjithPanneerselvam]

This PR implements a new function oldest(x[, count]) that returns the oldest count commits in x, based on committer timestamp and the default count is 1. This is an exact complementary function to latest() available in the revset language.

Here are a few use cases that I could think about

• oldest(author(martinvonz), 5) returns the oldest 5 commits from the author martinvonz.
• oldest(all(), 5) returns the oldest 5 commits in this repository.

Checklist

If applicable:

• I have updated CHANGELOG.md
• I have updated the documentation (README.md, docs/, demos/)
• I have updated the config schema (cli/src/config-schema.json)
• I have added tests to cover my changes

[PhilipMetzger]

Hello, welcome to the project. This can be all one commit and should be titled something like revset:, per https://github.com/jj-vcs/jj/blob/main/docs/contributing.md#code-reviews

[martinvonz]

nit: it seems like it should be lastest() and earliest() (or newest() and oldest()). Or maybe the inconsistency is fine.

[AjithPanneerselvam]

nit: it seems like it should be lastest() and earliest() (or newest() and oldest()). Or maybe the inconsistency is fine.

Yeah, it makes sense! I'll go with earliest to stay consistent with the existing latest function.

[emilazy]

I like the rename to newest()/oldest(), because currently people often use latest() when they want heads(). The renames seem less confusable with the topological notion that is usually more relevant.

[AjithPanneerselvam]

I like the rename to newest()/oldest(), because currently people often use latest() when they want heads(). The renames seem less confusable with the topological notion that is usually more relevant.

Ah, I just renamed the references in this PR to earliest() 😅 That's okay I can revert this commit.

@emilazy I reckon the renaming from latest() to newest() is a breaking change. Can I rename it in a separate PR?

cc: @martinvonz

[emilazy]

Yeah not trying to hold up this PR, just saying that I think renaming them both actually seems like a good idea in general. We’d want to keep a backwards compatibility alias etc. for a while.

[PhilipMetzger]

Sorry, it is the commit message which should contain revset: not the PR title.

[yuja]

I like the rename to newest()/oldest(), because currently people often use latest() when they want heads(). The renames seem less confusable with the topological notion that is usually more relevant.

fwiw, latest() could have some topological notion assuming that all ancestors commits are (usually) older than the commit itself. The current implementation scans all commits, which is inefficient.