Support custom preparationBatchSize defined via SourceKit's options

[josh-arnold-1]

What

Follow up from this issue here: #2238

Allow projects to define a custom preparationBatchSize in their projects. This can be incredibly useful for my large Bazel based project.

Currently, I've only supported a target strategy which allows you to define a constant batch size. We can easily extend this in the future for multiple strategies.

Usage

"preparationBatchingStrategy": {    
  "strategy": "target",
  "batchSize": 200    
}

Test plan

Tested locally in my project.

[ahoppen]

Thanks for picking this up @josh-arnold-1 🙏. One high-level comment: I would really like us to design the configuration option that allows us to customize the batching strategy as I described in #2238 (comment). do you think you could look into that?

[josh-arnold-1]

Thanks for picking this up @josh-arnold-1 🙏. One high-level comment: I would really like us to design the configuration option that allows us to customize the batching strategy as I described in #2238 (comment). do you think you could look into that?

Thanks for the review!

What if we update the schema to be an enum of configuration options like you specified, but we just supply a single option for now, which we default to a target size of 1 to maintain SourceKit-LSPs default behavior?

That way, we can easily extend additional strategies in future PRs, whilst maintaining the current configuration API?

What are your thoughts? Thanks!

      {
        "type": "object",
        "description": "Prepare a fixed number of targets in a single batch",
        "properties": {
          "strategy": {
            "const": "target"
          },
          "batchSize": {
            "type": "integer",
            "description": "Defines how many targets should be prepared in a single batch"
          }
        },
        "required": [
          "strategy",
          "batchSize"
        ]
      },

[ahoppen]

Your proposal for the JSON schema sounds great to me!

[brentleyjones]

Is there a way to set batchSize to inf/"batch everything at once"?

[bnbarham]

Is there a way to set batchSize to inf/"batch everything at once"?

Setting to a high value seems reasonable to me for this rather than handling it specifically

[josh-arnold-1]

Sorry for the delayed response — I was OOO recently. I’m looking into this now and realized that config.schema.json generation might not support enums with associated values (unless I’m missing something?).

If that’s the case, what would be the best way to represent the different batching strategies?

@ahoppen, any guidance here would be super helpful. Thanks!

[ahoppen]

Yeah, the JSON schema generation would need to be expanded to support this. That’s what I meant in #2238 (comment).

As a side note, allowing us to generate the oneOf in the schema above like this will likely need quite a bit new functionality in ConfigSchemaGen. If we only stick to the target-based strategy, we should only need support for the const key in the JSON schema, which should be a lot easier to accomplish.

[josh-arnold-1]

Thanks @ahoppen, I just updated the code with what we discussed!

[ahoppen]

Thank you. I really appreciate that you put in the effort to generate the enum options in the JSON schema and Markdown document 🙏🏽 Just a few nitpicky comments.

Two other high-level comments:

• Could you include the test from #2238 in this PR as well?
• I would also like to see the BSP server advertising if it can handle multi-target preparation as I mentioned in #2238 (comment) so that users can’t get regressed performance for SwiftPM projects by increasing the target batch size. #2238 already implements most of this, so you should be able to just copy it. I would be happy for this to be follow-up PR though so we can get this one in without any further discussions on the BSP protocol extension.

[josh-arnold-1]

Sorry for the month-long hiatus, I was quite strapped for time.

I tried to respond to all the comments, please let me know if I missed anything! Thanks @ahoppen for your time reviewing this 🙏

[ahoppen]

No worries with the delay and thanks again for your work on this. I have two nitpicky comments, but I’d be happy to merge this as-is as well and address those in follow-up PRs.

[ahoppen]

@swift-ci Please test

[ahoppen]

@swift-ci Please test Windows

[josh-arnold-1]

Thank you for the review! I responded to the comments. Happy thanksgiving 🦃

[ahoppen]

Thanks. LGTM. Very excited to see this get in 🤩

[ahoppen]

@swift-ci Please test

[josh-arnold-1]

@ahoppen just updated! Thank you!

[ahoppen]

@swift-ci Please test

[josh-arnold-1]

I see the testing failed — where is the right place to interpret the test failure logs? I can look into this and try to fix. Thanks!

[ahoppen]

Searching for the last occurrences of error: yields the following. I think you need to rebase your PR on top of main to pick up #2366 and then resolve the build failure.

[2025-12-03T03:55:36.906Z] /Users/ec2-user/jenkins/workspace/swift-sourcekit-lsp-PR-macOS/branch-main/sourcekit-lsp/Sources/SKOptions/PreparationBatchingStrategy.swift:46:34: error: use of protocol 'Encoder' as a type must be written 'any Encoder'; this will be an error in a future Swift language mode [#ExistentialAny]
[2025-12-03T03:55:36.906Z] 44 |   }
[2025-12-03T03:55:36.906Z] 45 | 
[2025-12-03T03:55:36.906Z] 46 |   public func encode(to encoder: Encoder) throws {
[2025-12-03T03:55:36.906Z]    |                                  `- error: use of protocol 'Encoder' as a type must be written 'any Encoder'; this will be an error in a future Swift language mode [#ExistentialAny]
[2025-12-03T03:55:36.906Z] 47 |     var container = encoder.container(keyedBy: CodingKeys.self)
[2025-12-03T03:55:36.906Z] 48 |     switch self {

[josh-arnold-1]

Updated! Thanks!

[ahoppen]

@swift-ci Please test

[ahoppen]

@swift-ci Please test Windows