Configuration

.exevra.yml must be a version-1 YAML object. The configuration file itself cannot be a symlink. Every baseline, reports, watched, and aggregation path is a non-empty relative path within the configuration root: absolute paths, .., backslashes, NUL bytes, and paths that resolve outside that root are rejected.

version: 1
baseline: .exevra/baseline.json
command: npm test -- --reporter=junit --outputFile=artifacts/junit.xml
reports:
  - artifacts/junit.xml
watched:
  - package.json
  - package-lock.json
policy:
  default:
    min_executed: 1
    max_drop_percent: 0
    identity: warn
    identity_details: counts
  protected_suites:
    - name: stable unit tests
      match: "^unit$"
      min_executed: 1
      max_drop_percent: 0
      identity: enforce
      identity_details: names

For a Maven root-only or multi-module project, init --maven adds this marker:

maven:
  modules: auto
  filter_policy: warn

version must be 1. baseline is the JSON baseline path. command is a non-empty Bash command. reports is a non-empty list of unique JUnit XML paths or filename patterns. A pattern may contain * in its filename, not its directory, and every matched XML file is checked. Literal report paths must be produced on every run. If all report entries are patterns, at least one must match; this lets init --maven work for Surefire-only, Failsafe-only, and combined Maven projects. watched is an optional list of paths matched as globs when a base ref is available.

policy.default is required. Its min_executed is a non-negative integer and max_drop_percent is a number from 0 through 100. identity accepts off, warn, or enforce and defaults to warn; identity_details accepts counts or names and defaults to counts.

policy.protected_suites is optional. Every item needs a non-empty name, a valid regular-expression match, and the same policy fields as default. Exevra selects the first protected-suite policy whose regular expression matches a suite name; otherwise it uses default.

When maven.modules is auto, Exevra follows literal <module> declarations from pom.xml files, recursively visits declared child POMs, and expands the configured standard Surefire/Failsafe report patterns in each non-aggregator module. A root-only project is represented by the same Maven marker and is checked as one module. A module may produce either report family or both. Modules without Maven test source files or current-build compiled test classes are not required to produce a report; a test-bearing module producing neither report family yields REPORT_MISSING. A matched report that cannot be read yields REPORT_UNREADABLE. Profiles, properties, parent inheritance, and custom report directories are not evaluated. Source detection recognizes common JVM test-source extensions under src/test; compiled .class files under target/test-classes from the current build also mark a module as test-bearing, and a direct non-profile testOutputDirectory declaration keeps it report-bearing.

maven.filter_policy controls lexical detection of test selection and skip flags in the configured command. It defaults to warn when omitted and accepts off, warn, or enforce. The seven supported flag families are -Dtest, -Dit.test, -Dgroups, -DexcludedGroups, -DskipTests, -Dmaven.test.skip, and -DskipITs. With warn, Exevra reports TEST_FILTERED and continues through report cleanup, command execution, and fresh-report evaluation. With enforce, it reports an error and stops after loading the configuration, before cleanup or execution. With off, it emits no filter finding.

The detector recognizes =value, a following value, and value-less flags where applicable at lexical token-like boundaries. Findings contain only the matched flag names; selector values and test names are never included. This is not complete Bash or Maven parsing and does not resolve shell expansion, aliases, profiles, properties, plugins, or custom Maven configuration. Non-Maven configurations are unchanged.

For Gradle multi-project builds, gradle.modules: auto reads literal project declarations from settings.gradle or settings.gradle.kts and expands each configured report path relative to every included project. Standard onboarding uses build/test-results/test/TEST-*.xml. Custom project directories assigned with project(...).projectDir = file(...) are supported. A missing or unreadable project report is retained as REPORT_MISSING or REPORT_UNREADABLE; a project with no settings file uses the root-only behavior. Maven and Gradle markers cannot be configured together.

gradle:
  modules: auto

Aggregation

The aggregation configuration and aggregate command are part of the 0.2.0 release.

aggregation is optional and is used only by aggregate. It needs a relative root, a non-empty list of unique shards, and a non-empty list of shard-relative JUnit reports. A shard is one directory name: it cannot contain a separator, ., .., or NUL.

aggregation:
  root: artifacts/shards
  shards:
    - unit-jdk17
    - unit-jdk21
  reports:
    - target/surefire-reports/TEST-*.xml

The collector expects each artifact under aggregation.root/<shard>, and applies each aggregation report pattern relative to that directory. The top-level reports remain the contract for check and record; aggregate does not run command, clear reports, or use watched paths. See Matrix and shard aggregation for the CI workflow.

Read Identity drift before using names.