Upstreaming the Scalar command

[dscho]

tl;dr: This series contributes the core part of the Scalar command to the Git project. This command provides a convenient way to clone/initialize very large repositories (think: monorepos).

Note: This patch series' focus is entirely on Scalar, on choosing sensible defaults and offering a delightful user experience around working with monorepos, and not about changing any existing paradigms for contrib/ (even if catching up on the mail thread is likely to give interested readers that false impression).

Changes since v9:

• The patches to build Scalar and run its tests as part of Git's CI/PR, have been dropped because a recent unrelated patch series does not interact well with them.

Changes since v8:

• The rebase on top of v2.34.0, which changed the default merge strategy to ORT, should have changed the default for merge.renames to true. This is now the case.
• Accommodate preemptively for ab/ci-updates which invalidates assumptions made by this patch series that would still hold true with v2.34.0 but are no longer valid in seen and would trigger CI build breakages.

Changes since v7:

• Clarified in the commit message why we cannot easily encapsulate the Scalar part of the CMake configuration in contrib/scalar/.
• Improved the README.md.

Changes since v6:

• Rebased on top of v2.34.0.
• Inserted a commit that adds contrib/scalar/README.md, containing the roadmap of what I have planned for Scalar.
• The Scalar test's definition of GIT_TEST_MAINT_SCHEDULER has been adjusted to accommodate for a change in v2.32.0..v2.34.0.
• The config setting defaults now include fetch.showForcedUpdates=false, which has been identified as helping with a performance issue in large repositories.
• To avoid mistaking the current patch series for being feature-complete enough to unleash onto end users, I moved the Makefile rules to build HTML/manual pages to a later patch series.
• The patch that adds support for -c <key>=<value> and -C <directory> was moved to its own add-on patch series: While it is obvious that those options are valuable to have, an open question is whether there are other "pre-command" options in git that would be useful, too, and I would like to postpone that discussion to that date.
• I added two patches that I had planned on keeping in an add-on patch series for later, to build and test Scalar as part of the CI. I am still not 100% certain that it is a good idea to do so already now, but let's see what the reviewers have to say.

Changes since v5:

• Fixed the commit message talking about make -C contrib/scalar/Makefile.
• Fixed the git ls-tree invocation suggested in the manual for scalar clone.
• Invoking make -C contrib/scalar, then changing a source file of libgit.a and then immediately invoking make -C contrib/scalar again will now implicitly rebuild libgit.a.

Changes since v4:

• scalar delete now refuses to delete anything if it was started from within the enlistment.
• scalar delete releases any handles to the object store before deleting the enlistment.
• The OBJECTS list in the Makefile will now include Scalar.
• scalar register now supports secondary worktrees, in addition to the primary worktree.

Changes since v3:

• Moved the "Changes since" section to the top, to make it easier to see what changed.
• Reworded the commit message of the first patch.
• Removed the [RFC] prefix because I did not hear any objections against putting this into contrib/.

Changes since v2:

• Adjusted the description of the list command in the manual page , as suggested by Bagas.
• Addressed two style nits in cmd_run().
• The documentation of git reconfigure -a was improved.

Changes since v1:

• A couple typos were fixed
• The code parsing the output of ls-remote was made more readable
• The indentation used in scalar.txt now consistently uses tabs
• We no longer hard-code core.bare = false when registering with Scalar

Background

Microsoft invested a lot of effort into scaling Git to the needs of the Windows operating system source code. Based on the experience of the first approach, VFS for Git, the Scalar project was started. Scalar specifically has as its core goal to funnel all improvements into core Git.

The present

The Scalar project provides a completely functional non-virtual experience for monorepos. But why stop there. The Scalar project was designed to be a self-destructing vehicle to allow those key concepts to be moved into core Git itself for the benefit of all. For example, partial clone, sparse-checkout, and scheduled background maintenance have already been upstreamed and removed from Scalar proper. This patch series provides a C-based implementation of the final remaining portions of the Scalar command. This will make it easier for users to experiment with the Scalar command. It will also make it substantially easier to experiment with moving functionality from Scalar into core Git, while maintaining backwards-compatibility for existing Scalar users.

The C-based Scalar has been shipped to Scalar users, and can be tested by any interested reader: https://github.com/microsoft/git/releases/ (it offers a Git for Windows installer, a macOS package and an Ubuntu package, Scalar has been included since v2.33.0.vfs.0.0).

Next steps

Since there are existing Scalar users, I want to ensure backwards-compatibility with its existing command-line interface. Keeping that in mind, everything in this series is up for discussion.

I obviously believe that Scalar brings a huge benefit, and think that it would be ideal for all of Scalar's learnings to end up in git clone/git init/git maintenance eventually. It is also conceivable, however, that the scalar command could graduate to be a core part of Git at some stage in the future (such a decision would probably depend highly on users' feedback). See also the discussion about the architecture of Scalar, kicked off by Stolee.

On top of this patch series, I have lined up a few more:

1. Implement a scalar diagnose command.
2. Use the built-in FSMonitor (that patch series obviously needs to wait for FSMonitor to be integrated).
3. Modify the config machinery to be more generous about concurrent writes, say, to the user-wide config.
4. A few patches to optionally build and install scalar as part of a regular Git install (also teaching git help scalar to find the Scalar documentation

These are included in my vfs-with-scalar branch thicket. On top of that, this branch thicket also includes patches I do not plan on upstreaming, mainly because they are too specific either to VFS for Git, or they support Azure Repos (which does not offer partial clones but speaks the GVFS protocol, which can be used to emulate partial clones).

One other thing is very interesting about that vfs-with-scalar branch thicket: it contains a GitHub workflow which will run Scalar's quite extensive Functional Tests suite. This test suite is quite comprehensive and caught us a lot of bugs in the past, not only in the Scalar code, but also core Git.

Epilogue

Now, to address some questions that I imagine every reader has who made it this far:

• Why not put the Scalar functionality directly into core Git, even a built-in? I wanted to provide an easy way for Git contributors to "play with" Scalar, without forcing a new top-level command into Git.
• Why implement the Scalar command in the Git code base? Apart from simplifying Scalar maintenance in the Microsoft port of Git, the tight version coupling between Git and Scalar reduces the maintenance burden even further. Besides, I believe that it will make it much easier to shift functionality from Scalar into core Git, once we took the hurdle of accepting the Scalar code into the code base.
• Why contribute Scalar to the Git project? We are biased, of course, yet our data-driven approach provides evidence that Scalar helps handling huge repositories with ease. By contributing it to the core Git project, we are able to share it with more users, especially some users who do not want to install Microsoft's fork of Git. We also hope that a lot of Scalar (maybe all of it) will end up in core Git, to benefit even more users.

cc: Derrick Stolee stolee@gmail.com
cc: Eric Sunshine sunshine@sunshineco.com
cc: Ævar Arnfjörð Bjarmason avarab@gmail.com
cc: Elijah Newren newren@gmail.com
cc: Bagas Sanjaya bagasdotme@gmail.com
cc: "Theodore Ts'o" tytso@mit.edu
cc: Matt Rogers mattr94@gmail.com
cc: Jeff King peff@peff.net