Skip to content

feat: Add the override marker to the flag and segment models and mark evaluations - #221

Draft
kinyoklion wants to merge 3 commits into
rlamb/overrides-java-filedata-reloaderfrom
rlamb/overrides-java-model-marker
Draft

kinyoklion wants to merge 3 commits into
rlamb/overrides-java-filedata-reloaderfrom
rlamb/overrides-java-model-marker

Conversation

@kinyoklion

@kinyoklion kinyoklion commented Sep 25, 2026 •

Copy link
Copy Markdown
Member

Summary

The OVERRIDE specification requires an evaluation to be marked as override-affected when any definition it read carried the override marker: the evaluated flag, a prerequisite at any depth, or a segment consulted during matching, whether or not the segment matched. The marking propagates upward only: a prerequisite's own record reflects only the definitions that its own subtree read, while the evaluation that requested it accumulates the result. An evaluation that fails is still marked when it read a marked definition, and a requested-type mismatch keeps the marking on its error reason, as does the WRONG_TYPE error that migrationVariation returns for an unrecognized stage.

  • DataModel.FeatureFlag and DataModel.Segment carry a transient isOverride marker. It is never serialized and never read from JSON. markedAsOverride() returns a shallow copy with the marker set, sharing nested collections and preprocessing with the source and never modifying it. The override layer will use it in the next change.
  • The evaluator tracks the marking per evaluation scope. The scope starts from its own flag's marker, each segment read can set it, and the value is saved, reset to the prerequisite's own marker, and restored around every prerequisite evaluation. Each result and each prerequisite record carries the marking on its reason.
  • EvalResult.isOverrideAffected() reads the reason's indicator, and withOverrideAffected(boolean) returns a copy so that the shared precomputed results are never mutated.
  • An evaluation that throws returns an EXCEPTION result marked with the evaluated flag's own marker, because that definition was read before the failure.
  • The server SDK now depends on launchdarkly-java-sdk-common 2.6.0, which adds EvaluationReason.isOverrideAffected() and the overrideAffected JSON property (written only when true). CI for this change cannot pass until that version is released.

Flag overrides are currently experimental and subject to change.

This PR is based on the file data reliability branch (rlamb/overrides-java-filedata-reloader) so the series applies in order; retarget it to feat/overrides once that branch merges.

The existing file data source keeps its current behavior; this change does not touch its code paths.

SDK-3246


Note

Overview
Adds experimental override tracking so evaluations can report when they used SDK-local flag or segment definitions instead of LaunchDarkly data.

Data model: FeatureFlag and Segment get a transient isOverride flag (not in JSON) plus markedAsOverride() shallow copies for the future override layer. Dependency: launchdarkly-java-sdk-common is bumped to 2.6.0 for EvaluationReason.isOverrideAffected() / serialization.

Evaluation: The evaluator accumulates override-affected state from the evaluated flag, prerequisite subtrees (scoped per prereq, upward propagation only), and any segment read during matching—even on non-match or errors. Results expose EvalResult.isOverrideAffected() and withOverrideAffected() without mutating shared precomputed instances. Wrappers (InputValidatingEvaluator, MigrationStageEnforcingEvaluator) preserve marking on type errors, exceptions, and allFlagsState reasons.

Reviewed by Cursor Bugbot for commit 64740d7. Bugbot is set up for automated code reviews on this repo. Configure here.

Opened from termag

@kinyoklion
kinyoklion force-pushed the rlamb/overrides-java-filedata-reloader branch from f0038ed to 79250a7 Compare September 28, 2026 21:03
@kinyoklion
kinyoklion force-pushed the rlamb/overrides-java-model-marker branch 3 times, most recently from 1ba722a to 996b8d3 Compare October 3, 2026 00:58
@kinyoklion

Copy link
Copy Markdown
Member Author

bugbot review

@cursor cursor Bot left a comment •

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Stale Bugbot comment from a previous run.

… evaluations

The OVERRIDE specification requires an evaluation to be marked as
override-affected when any definition it read carried the override
marker: the evaluated flag, a prerequisite at any depth, or a segment
consulted during matching, whether or not the segment matched. The
marking propagates upward only. A prerequisite's own record reflects
only the definitions that its own subtree read. An evaluation that fails
is still marked when it read a marked definition.

DataModel.FeatureFlag and DataModel.Segment carry a transient
isOverride marker that is never serialized and never read from JSON,
with a markedAsOverride() shallow copy that the override layer will use
without modifying the source's entity.

The evaluator tracks the marking per evaluation scope: the scope starts
from its own flag's marker, each segment read can set it, and the value
is saved, reset to the prerequisite's marker, and restored around each
prerequisite evaluation. The result and every prerequisite record carry
the marking on their reason. EvalResult exposes isOverrideAffected() and
withOverrideAffected(), which copies rather than mutating the shared
precomputed results. A requested-type mismatch keeps the marking on its
error reason.

The server SDK now depends on launchdarkly-java-sdk-common 2.6.0, which
adds the overrideAffected indicator to EvaluationReason. CI for this
change cannot pass until that version is released.

Flag overrides are currently experimental and subject to change.
@kinyoklion
kinyoklion force-pushed the rlamb/overrides-java-model-marker branch from 996b8d3 to ca22bc0 Compare October 13, 2026 21:02
…gnized stage

migrationVariation replaces the result of a flag that serves an
unrecognized stage with a WRONG_TYPE error. The evaluation read the same
definitions, so the replacement keeps the override-affected marking, as
the type mismatch of an ordinary variation call already does, and the
migration tracker receives a marked reason for an overridden flag.
@kinyoklion

Copy link
Copy Markdown
Member Author

bugbot review

@cursor cursor Bot left a comment •

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Stale Bugbot comment from a previous run.

When the evaluation of a flag throws, the error result now carries the
override-affected marking of the flag's own definition, which was read
before the failure. The spec marks a failed evaluation when a definition
it read came from the override store.
@kinyoklion

Copy link
Copy Markdown
Member Author

bugbot review

@cursor cursor Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

✅ Bugbot reviewed your changes and found no new issues!

Comment @cursor review or bugbot run to trigger another review on this PR

Reviewed by Cursor Bugbot for commit 64740d7. Configure here.

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant