Striff logo

Public pull requests

Four changes, and what the structure said about them.

Every example below is a real pull request in a public repository. The diagram is the one Striff drew; the boxed text is what it produced, word for word. Under each one is the check we ran by hand against both revisions before putting it here — so you can repeat it rather than take our word for it.

Homepage gallery Install extension
What you are looking at

A structural diff: the components a change touched, and the dependencies between them at both revisions. Additions in green, removals struck through, and the relationships that only exist on one side of the change.

Why these four

Each shows something the file diff cannot: a type's dependents, a loop that closes, a module boundary crossed by two files, a sentence in a design doc checked against the code.

How often this fires

These four were picked from 79 public pull requests across 27 repositories. 45 of them produced no finding at all. Most pull requests do not move the architecture, and Striff is built to say so quietly.

The reviews

Four public pull requests.

Open any diagram and drag it. Open the pull request beside it and read the diff. The two should agree.

Contract change smallrye/smallrye-graphql #1424 Open PR

#1423: add GraphQLResult and wrap successful result

Two accessors on a schema model class were renamed. The rename is one line in the diff. The number of components that depend on the class is not in the diff at all.
Structural diff 1 structural regression detected
base 6e20902 head ee8436b
Striff structural diff for smallrye/smallrye-graphql #1424
Scroll to zoom · Drag to pan

What Striff produced

HIGH Breaking contract change in Wrapper
Wrapper removed public methods isNotEmpty() and setNotEmpty(boolean). 12 components depend on this type, and the removal may cause compilation or runtime failures in those dependents. This is a stable contract change that should be coordinated across the codebase.

Suggested direction Restore the removed methods isNotEmpty() and setNotEmpty(boolean) in Wrapper and deprecate them instead, or update all 12 known dependents (e.g., AbstractHelper, DefaultMapAdapter, DataFetcherFactory, Bootstrap) to use the new isNonNull()/setNonNull(boolean) equivalents before removal.

What we checked

  • Wrapper.java, in the common/schema-model module, declares setNotEmpty(boolean) and isNotEmpty() at the base revision. At the head revision it declares setNonNull(boolean) and isNonNull() instead — and neither of the two methods the check names.
  • Twelve files under src/main reference Wrapper at both revisions — the same count the check reports.
  • All four dependents the suggested fix names — AbstractHelper, DefaultMapAdapter, DataFetcherFactory, Bootstrap — reference Wrapper at the head revision.
  • The rename is the pull request's own work: the commit that performs it is on the branch, and the base revision is the merge base, so nothing here is an artefact of the branch being behind.
  • Only two isNotEmpty() calls survive anywhere in the repository at the head revision, and both are AssertJ string assertions in integration tests that have nothing to do with this class.
Open the pull request Compare 6e20902 … ee8436b
A loop closed citrusframework/citrus-simulator #370 Open PR

feat: persist only failed scenarios

A listener class gained a constructor parameter so it could read a configuration flag. That one parameter is the last edge of a loop that runs through four packages.
Structural diff 7 components, 13 relationships, no items surfaced (2 evidence-only findings)
base c658ef8 head 51917ce
Striff structural diff for citrusframework/citrus-simulator #370
Scroll to zoom · Drag to pan

What Striff produced

LOW New directional dependency
New dependency from org.citrusframework.simulator.listener to org.citrusframework.simulator.config, with no prior edge in this direction.
MEDIUM Package cycle detected
A dependency cycle among 11 packages (37 edges) already turned before this PR, which adds org.citrusframework.simulator.listener -> org.citrusframework.simulator.config to it.

Both of these were recorded as evidence behind the diagram. Neither was raised as an item on the check — the loop was already turning, and this change adds an edge to it rather than creating it.

What we checked

  • At the base revision, no file in org.citrusframework.simulator.listener references org.citrusframework.simulator.config. At the head revision, SimulatorStatusListener imports SimulatorConfigurationProperties.
  • The way back was already there, in production sources only, at the base revision: SimulatorConfigurer (config) imports ScenarioMapper (scenario.mapper); ScenarioMappers imports HttpOperationScenario (http); SimulatorRestAutoConfiguration imports SimulatorMessageListener (listener).
  • Those three edges plus the one this pull request adds form a closed loop: config → scenario.mapper → http → listener → config.
  • Rebuilding the package graph from imports independently puts both listener and config inside one strongly connected component at the base revision, which is what the second finding says: the cycle predates the change.
Open the pull request Compare c658ef8 … 51917ce
Boundary crossing smallrye/smallrye-graphql #2083 Open PR

Add custom scalar support to client value formatter

Two files changed. One of them is a formatting helper inside the client, and it started importing types that are declared in the server module.
Structural diff 11 components, 13 relationships, no items surfaced (1 evidence-only finding)
base c48e433 head b2484d4
Striff structural diff for smallrye/smallrye-graphql #2083
Scroll to zoom · Drag to pan

What Striff produced

LOW New directional dependency
New dependency from io.smallrye.graphql.client.impl.core.utils to io.smallrye.graphql.api, with no prior edge in this direction.

This is the quiet end of the output. One evidence-only finding, nothing raised on the check, and a diagram that puts the new edge in front of you without anybody having to argue about it.

What we checked

  • The pull request changes exactly two files: ValueFormatter.java and its test.
  • At the base revision the package io.smallrye.graphql.client.impl.core.utils holds one production file and one test file, and neither mentions io.smallrye.graphql.api.
  • At the head revision ValueFormatter imports CustomFloatScalar, CustomIntScalar and CustomStringScalar, and branches on all three when formatting a value.
  • Those three types are declared under server/api/src/main/java/io/smallrye/graphql/api/ — a different Maven module from the client implementation that now reaches for them.
Open the pull request Compare c48e433 … b2484d4
Documented rules spring-ai-community/spring-ai-session #33 Open PR

Add runtime branch isolation to SessionMemoryAdvisor

Nineteen sentences from this repository's own docs folder became checkable rules on this pull request. All nineteen held. Six of them are below, with the sentence each one came from.
Structural diff 2 components, 2 relationships, no structural anomalies
base b283ba5 head 0e70aa2
Striff structural diff for spring-ai-community/spring-ai-session #33
Scroll to zoom · Drag to pan

Rules read out of the docs

  1. docs/session-management/multi-agent.md HOLDS
    To apply branch isolation automatically inside SessionMemoryAdvisor, configure the eventFilter on the builder:

    SessionMemoryAdvisor depends on EventFilter

    SessionMemoryAdvisor.java imports EventFilter and holds one as a field

  2. docs/session-management/cross-session-recall.md HOLDS
    The term count is capped at 20 per call (CrossSessionRecallTools.MAX_QUERY_TERMS).

    CrossSessionRecallTools declares MAX_QUERY_TERMS as a field

    CrossSessionRecallTools.java line 75, `static final int MAX_QUERY_TERMS = 20`

  3. docs/session-management/compaction.md HOLDS
    All four strategies share a common safety rule enforced by CompactionUtils.snapToTurnStart:

    CompactionUtils declares snapToTurnStart as a method

    CompactionUtils.java line 99

  4. docs/session-management/recall-storage.md HOLDS
    Events dropped from the active window are flagged SessionEvent.isArchived() and kept in the log.

    SessionEvent declares isArchived as a method

    SessionEvent.java line 153

  5. docs/session-management/recall-storage.md HOLDS
    Page size defaults to EventFilter.DEFAULT_PAGE_SIZE (10) and is configurable via the builder (SessionEventTools.builder(sessionService).pageSize(20).build()).

    EventFilter declares DEFAULT_PAGE_SIZE as a field, and SessionEventTools declares builder as a method

    EventFilter.java line 133 (`= 10`); SessionEventTools.java line 87

  6. docs/session-management/compaction.md HOLDS
    This ensures tokensEstimatedSaved in CompactionResult accurately reflects the full cost of removed events, including tool-heavy turns.

    CompactionResult declares tokensEstimatedSaved as a field

    CompactionResult.java line 31, a component of the record

Nineteen rules, nineteen held, nothing raised. That is the ordinary result, and a page that only showed you the other kind would be lying about how often this fires.

What we checked

  • Every sentence above appears verbatim in the file it is attributed to, at the head revision, once whitespace from the Markdown line wrapping is normalised.
  • Every declaration each rule asserts was read by hand in the source at the head revision; the file and line are printed beside it.
  • Every quote is prose from the documentation, not a code sample lifted out of a fenced block.
  • The base revision is the merge base of the pull request, so the comparison is the branch's own work and nothing else.
Open the pull request Compare b283ba5 … 0e70aa2

These are other people's repositories. Nothing here is a judgement about the code or the people who wrote it — it is a record of what a change did to the shape of a codebase, and every one of these pull requests is a perfectly reasonable change to make.