[magnusp]

Why

The URL mapping section said the query string is ignored when deriving the namespace. That contradicts XerialConformanceDriver.plan(), its class Javadoc, and two harness tests. A conformance author reading the README would treat two spellings of the same file as one database when they are isolated databases.

What changed

• The URL mapping paragraph now says the query string is part of the namespace identity.
• It keeps the correct fact that the option values are not applied to the server.
• The "ignored" wording is gone.

Scope

Docs only. No code or test changes. The canonical statement work in #171 and #174 is untouched.

Blast Radius

Only the conformance README. No runtime behavior changes.

Verification

• XerialHarnessTest#queryStringIsPartOfTheFileNamespace: Tests run: 1, Failures: 0, Errors: 0. ?mode=ro and ?mode=rw derive different namespaces.
• XerialHarnessTest#fileNamespacesWithDifferentQueryStringsAreIsolated: Tests run: 1, Failures: 0, Errors: 0. The two variants cannot see each other's tables.
• ./mvnw spotless:check exits 0.

Closes #182.

[magnusp]

Verdict: PASS

Revision verified

• Head SHA: a0a408e416c19e06e8f298acdacfe4e99eb17fed
• Base SHA: fc3f819282cbeff71accb34b7a006b6e90400e81 (base branch main)
• Patch-id (stable): fe92fd7c29721867157ead61218ce08db213c5e6
• Diff: 1 file changed, +4 / -2, libsql-conformance/README.md only

Acceptance criteria (issue #182)

1. README no longer says the query string is ignored — MET. grep -rn "query string is ignored\|ignored when deriving" --include=*.md returns nothing repo-wide; the only "query string" mentions left are the new paragraph.
2. Paragraph states both facts — MET. New §URL mapping text: "The query string is part of the namespace identity, so two URLs that differ only in their query string address different namespaces. The option values themselves are not applied to the conformance server; a URL pragma such as ?journal_mode=WAL selects a separate namespace rather than changing server behavior." The first fact matches plan() (identity = normalized + "?" + query, hashed into the namespace); the second fact (option values not applied) is retained.
3. Wording consistent with the XerialConformanceDriver class Javadoc — MET. Javadoc: "The query string is part of the namespace identity, so two URLs with different options address different namespaces; the options themselves are not applied to the server." The README restates the same two claims with no contradiction.

Behavior verified from the head checkout (commands exactly as requested)

1. mise exec -- ./mvnw -q -pl libsql-conformance -am -Dit.test='XerialHarnessTest#queryStringIsPartOfTheFileNamespace' -Dfailsafe.failIfNoSpecifiedTests=false verify → exit 0; report line: Tests run: 1, Failures: 0, Errors: 0, Skipped: 0
2. mise exec -- ./mvnw -q -pl libsql-conformance -am -Dit.test='XerialHarnessTest#fileNamespacesWithDifferentQueryStringsAreIsolated' -Dfailsafe.failIfNoSpecifiedTests=false verify → exit 0; report line: Tests run: 1, Failures: 0, Errors: 0, Skipped: 0

Reviewer notes

• The second test genuinely exercised sqld, not a short-circuit. It starts the container lazily, but the JUnit LauncherSessionListener guard (LibsqlConformanceGuard.verify()) opens a connection before test execution, which starts the container. That is why the reported testcase time is only ~0.06-0.1 s while total Maven wall time was ~9 s. The guard fails the run unless the connection reports the libSQL driver/product and an alt1 source id; the run used ghcr.io/tursodatabase/libsql-server:v0.24.33 and reported Skipped: 0.
• Both tests are pre-existing. This PR changes docs only, so they confirm the documented behavior rather than the diff itself.
• Non-blocking, outside Fix libsql-conformance README: query string is part of the namespace identity #182's scope: the paragraph does not call out the file:...?mode=memory special case, where the URL maps to a fresh in-memory namespace per connection (retained=false). It does not contradict the new wording.