Honor PropertyNamingStrategy for is-prefixed record components

[seonwooj0810]

Problem

When a Jackson PropertyNamingStrategy (e.g. SNAKE_CASE) is configured on the ObjectMapper used by ModelResolver, non-boolean record components whose name starts with is are left in camelCase in the generated schema, while every other field is translated correctly.

Given:

public record PidLookupResponse(
        String familyName, String expiryDate,
        String issuanceDate, String issuingCountry, String issuingAuthority) {}

with SNAKE_CASE, the resolved schema contains family_name, expiry_date — but issuanceDate, issuingCountry, issuingAuthority (untranslated).

This was diagnosed as a swagger-core issue by the springdoc-openapi maintainer in springdoc/springdoc-openapi#3293 (reproducible directly against ModelResolver, no Spring involved).

Root cause

The get/is name-clobbering hack (added for #415) in ModelResolver falls back to the raw member name when an accessor starts with get/is followed by a lowercase letter:

if (altName.startsWith(prefix) && length > offset
        && !Character.isUpperCase(altName.charAt(offset))) {
    propName = altName;
}

For a record, the accessor name equals the component name (issuanceDate), so "is" + lowercase matches and propName is overwritten with the camelCase member name — discarding the strategy-translated value that Jackson already computed in propDef.getName(). Classic getters (getIssuanceDate/isActive) are unaffected because the character after the prefix is uppercase.

Fix

Skip the hack when a PropertyNamingStrategy is configured. In that case propDef.getName() is already authoritative and must not be overridden. Default (no-strategy) behavior is unchanged, so the #415 workaround is fully preserved.

if (propDef.getPrimaryMember() != null
        && _mapper.getSerializationConfig().getPropertyNamingStrategy() == null) {

Tests

Added Ticket3293Test in the swagger-java17-support module (records require Java 17). It resolves the record above with a SNAKE_CASE ModelResolver and asserts all five string components plus a genuine boolean component (isActive → is_active) are snake_cased.

Verified the test fails without the fix (the three is* components leak as issuanceDate/issuingCountry/issuingAuthority) and passes with it, with is_active correct in both runs (no boolean-getter regression). Full swagger-java17-support module and the io.swagger.v3.core.resolving.* package pass.

Refs springdoc/springdoc-openapi#3293

[Mattias-Sehlstedt]

Hi, first I would like to clarify that I am not the maintainer of springdoc, only a contributor. So take everything I say with a grain of salt.

Second, this looks like to be a duplicate of your previous PR #5192?

[seonwooj0810]

You're right, thanks for catching this @Mattias-Sehlstedt. This does overlap with my earlier #5192, which targets the same ModelResolver get/is name-clobbering hack but more broadly (both get- and is-prefixed names, not just is-prefixed record components). To avoid two competing PRs for the same fix I'll close this one in favor of #5192. Apologies for the duplicate.

[seonwooj0810]

Closing as a duplicate of #5192, which addresses the same root cause more comprehensively.