fix: do not charge actual cost for denied fields

[ysmolski]

Do not charge actual costs when some feld is denied or
nulled because some of its non-nullable children was null.

We collect stats on render pass in Resolvable too.
This gives easy way to handle stats for nulled items of arrays.

[coderabbitai]

📝 Walkthrough

Walkthrough

Adds authorization-aware actual cost accounting: cost paths move to rendered fieldPath, resolution records cost stats after render-time authorization/null propagation, ComputeActualCost uses resolve context stats, execution gains authorizer options, and tests cover denied-field cost behavior.

Changes

Auth-aware cost accounting

Layer / File(s)	Summary
CostTreeNode fieldPath rename and lookups v2/pkg/engine/plan/cost.go, v2/pkg/engine/plan/cost_visitor.go	CostTreeNode stores fieldPath instead of jsonPath; cost lookup logic uses fieldPath for runtime type stats, abstract-type weighting, list multipliers, unreached-subtree pruning, AST path construction, and debug output.
Render-time rendered-field-path cost tracking v2/pkg/engine/resolve/resolvable.go	Resolvable records cost-control stats during render using renderFieldPath(); adds reachedFields memoization, centralizes authorization-coordinate derivation, and updates array/object stat collection and reset/init behavior.
ComputeActualCost signature and execution options execution/graphql/request.go, execution/engine/execution_engine.go	Request.ComputeActualCost now accepts *resolve.Context; the execution engine passes the resolve context and adds authorizer options for post-fetch and pre-fetch field authorization.
Denied-field cost tests execution/engine/execution_engine_cost_auth_test.go	Adds authorizers and TestExecutionEngine_Cost_DeniedFields subtests covering control cases, pre-fetch and post-fetch denials, non-nullability propagation, and list behavior.

Estimated code review effort: 4 (Complex) | ~60 minutes

Possibly related PRs

• wundergraph/graphql-go-tools#1505: Both PRs modify how ExecutionEngine.Execute invokes Request.ComputeActualCost with different resolve context inputs.
• wundergraph/graphql-go-tools#1512: Both PRs address actual cost computation and test assertions in the execution engine under runtime behavior changes.
• wundergraph/graphql-go-tools#1542: Both PRs modify v2/pkg/engine/plan/cost.go affecting how actual/estimated costs are computed.

🚥 Pre-merge checks | ✅ 4

✅ Passed checks (4 passed)

Check name	Status	Explanation
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.
Title check	✅ Passed	The title clearly matches the main change: preventing actual cost charges for denied fields.
Description check	✅ Passed	The description is directly related to the changeset and describes the new actual-cost and render-pass stats behavior.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch yury/eng-9855-do-not-charge-actual-cost-for-denied-fields

---

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

[coderabbitai]

Actionable comments posted: 2

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@execution/graphql/request.go`:
- Around line 211-218: Guard resolveCtx before using it in
Request.ComputeActualCost so nil callers don’t panic; keep the existing fallback
behavior of setting actualCost to 0 when cost cannot be computed. Update the
initial nil check in ComputeActualCost to include resolveCtx alongside calc and
resolveCtx.TypeNameStats, and only call calc.ActualCostWithDenials when
resolveCtx is non-nil and has the required stats.

In `@v2/pkg/engine/plan/cost.go`:
- Around line 968-972: The comment on ActualCostWithDenials currently describes
typeNameDenials as JSON paths, but the matching logic in isFieldDenied uses
exact resolver field paths including the leading operation type. Update the
documentation in ActualCostWithDenials and any nearby references to say these
are resolver field paths, not JSON paths, and clarify the required format with
the operation type prefix so callers use the same keys that isFieldDenied
expects.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro

Run ID: dfa4049f-399c-4327-b4a3-a842df9dc21f

📥 Commits

Reviewing files that changed from the base of the PR and between 66ef719 and ac3e2ee.

📒 Files selected for processing (8)

• execution/engine/execution_engine.go
• execution/engine/execution_engine_cost_auth_test.go
• execution/graphql/request.go
• v2/pkg/engine/plan/cost.go
• v2/pkg/engine/plan/cost_visitor.go
• v2/pkg/engine/resolve/context.go
• v2/pkg/engine/resolve/resolvable.go
• v2/pkg/engine/resolve/resolve.go

[coderabbitai]

🧹 Nitpick comments (1)

execution/engine/execution_engine_cost_auth_test.go (1)

263-284: 🎯 Functional Correctness | 🔵 Trivial | 💤 Low value

Rename the stale test titles
expectedActualCost is 5 in both cases, so only Query.user is charged. Update the titles to match that behavior:

• denied non-nullable field nulls the parent: earlier sibling is charged → ... earlier sibling is also not charged
• subgraph null for non-nullable field nulls the parent: earlier sibling and the field are charged → ... neither sibling nor the field is charged

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@execution/engine/execution_engine_cost_auth_test.go` around lines 263 - 284,
The affected subtests in execution_engine_cost_auth_test.go have stale names
that no longer match the asserted cost behavior. Update the `t.Run(...)` titles
around the `expectedActualCost` assertions so they describe that only
`Query.user` is charged, using the same test cases and identifiers (`denied
non-nullable field nulls the parent` and `subgraph null for non-nullable field
nulls the parent`) to rename them to the corrected “earlier sibling is also not
charged” and “neither sibling nor the field is charged” wording.

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Nitpick comments:
In `@execution/engine/execution_engine_cost_auth_test.go`:
- Around line 263-284: The affected subtests in
execution_engine_cost_auth_test.go have stale names that no longer match the
asserted cost behavior. Update the `t.Run(...)` titles around the
`expectedActualCost` assertions so they describe that only `Query.user` is
charged, using the same test cases and identifiers (`denied non-nullable field
nulls the parent` and `subgraph null for non-nullable field nulls the parent`)
to rename them to the corrected “earlier sibling is also not charged” and
“neither sibling nor the field is charged” wording.

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro

Run ID: c844cd03-fbab-4cef-8e47-96d65598ed53

📥 Commits

Reviewing files that changed from the base of the PR and between 39454aa and 8b2e1b5.

📒 Files selected for processing (5)

• execution/engine/execution_engine_cost_auth_test.go
• execution/graphql/request.go
• v2/pkg/engine/plan/cost.go
• v2/pkg/engine/plan/cost_visitor.go
• v2/pkg/engine/resolve/resolvable.go

🚧 Files skipped from review as they are similar to previous changes (1)

• v2/pkg/engine/plan/cost_visitor.go