fix: calculate costs for abstract fields without double counting

[ysmolski]

This PR fixes how costs for abstract types are calculated. Especially for fragments used on abstract types.

For example, for actual costs, instead of summing each type of the union, engine sums only costs for types actually seen in the returned results.

For estimated costs, instead of summing all the implementing nodes of abstract types, we just pick the implementing type with the maximum cost and add that to the costs of fields selected on the abstract type itself.

I had to recalculate and verify numbers in tests.

Unfortunately, this fix is not final. There is unfinished work with how the cost of the field itself is selected. Right now we pick the maximum weight across implementing types and dataSources independently from children costs selection. This concerns the estimation the most. The better approach would be to be tie this process with the selection of the type with maximum cost.

[claude]

Claude Code Review

This repository is configured for manual code reviews. Comment @claude review to trigger a review and subscribe this PR to future pushes, or @claude review once for a one-time review.

_{Tip: disable this comment in your organization's Code Review settings.}

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro

Run ID: 1fbb71ed-50c0-4db0-a8ec-2fd2b6e5feaa

📥 Commits

Reviewing files that changed from the base of the PR and between e6ace3c and 8f673b4.

📒 Files selected for processing (3)

• execution/engine/execution_engine_cost_test.go
• execution/graphql/request.go
• v2/pkg/engine/plan/cost.go

🚧 Files skipped from review as they are similar to previous changes (3)

• execution/graphql/request.go
• v2/pkg/engine/plan/cost.go
• execution/engine/execution_engine_cost_test.go

---

📝 Walkthrough

Walkthrough

This PR replaces integer list-size tracking with TypeNameStats (Size + per-typename counts). The resolver records per-path TypeNameStats, Context exposes TypeNameStats, cost computation uses TypeNameStats for actual multipliers, call sites are updated, and tests are re-baselined.

Changes

TypeNameStats Migration

Layer / File(s)	Summary
TypeNameStats model and runtime collection v2/pkg/engine/resolve/resolvable.go	Introduce TypeNameStats and collect per-path Size and __typename distributions during array/object pre-walk.
Context and resolver propagation v2/pkg/engine/resolve/context.go, v2/pkg/engine/resolve/resolve.go	Replace ActualListSizes with TypeNameStats on Context; resolver assigns resolvable.typeNameStats into ctx.TypeNameStats and Free clears it.
Cost defaults and list-size initialization v2/pkg/engine/plan/cost.go	Add unexported default weights and sentinels; initialize FieldListSize.multiplier and fallback behavior for list sizing.
Core cost computation and abstract handling v2/pkg/engine/plan/cost.go	Introduce costNodeResult; refactor node cost aggregation, field/arg/directive accumulation, and switch abstract-child aggregation and multiplier logic to use TypeNameStats in actual mode.
Cost public API and debug printing v2/pkg/engine/plan/cost.go	Change CostCalculator.ActualCost/DebugPrint to accept map[string]resolve.TypeNameStats and emit structured per-node component debug output.
Execution/request wiring execution/graphql/request.go, execution/engine/execution_engine.go	Request.ComputeActualCost and synchronous ExecutionEngine.Execute now pass TypeNameStats into cost calculation.
Tests, fixtures, and rebaselines execution/engine/execution_engine_cost_test.go, execution/engine/execution_engine_test.go	Re-baseline many expected estimated/actual costs, add a comprehensive abstract-list concrete-fragment test suite, update fixtures to include concrete __typename entries, and add t.Parallel() in several subtests.
Object abstractness helper v2/pkg/engine/resolve/node_object.go	Add Object.isAbstract() helper to detect abstract-returning object semantics used by resolver/type-stat collection.

Estimated code review effort

🎯 4 (Complex) | ⏱️ ~60 minutes

Possibly related PRs

• wundergraph/graphql-go-tools#1512: Updates to test expectations for actual costs intersect with this PR’s changes to actual-cost inputs and test assertions.
• wundergraph/graphql-go-tools#1493: Related adjustments to actual cost list-multiplier/denominator logic in v2/pkg/engine/plan/cost.go.
• wundergraph/graphql-go-tools#1502: Related empty-list and list-multiplier actual-cost behavior updates.

🚥 Pre-merge checks | ✅ 4

✅ Passed checks (4 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title 'fix: calculate costs for abstract fields without double counting' accurately and specifically describes the main change in the PR, which refactors cost calculation for abstract GraphQL types.
Description check	✅ Passed	The description clearly explains how costs for abstract types are calculated differently for actual and estimated costs, with specific examples and acknowledgment of incomplete work.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch yury/eng-9581-inline-fragments-on-abstract-list-types-charge-all-items

---

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[ysmolski]

@coderabbitai review

[coderabbitai]

✅ Actions performed

Review triggered.

Note: CodeRabbit is an incremental review system and does not re-review already reviewed commits. This command is applicable only when automatic reviews are paused.

[ysmolski]

@claude review once

[claude]

⚠️ Code review skipped — your organization's overage spend limit has been reached.

Code review is billed via overage credits. To resume reviews, an organization admin can raise the monthly limit at claude.ai/admin-settings/claude-code.

Once credits are available, comment @claude review on this pull request to trigger a review.