fix: expand Array arguments in MarkerLoggingAdapter log templates (#3257)

[He-Pin]

Motivation

MarkerLoggingAdapter.format1 passed an Array argument to the varargs
format(t, arg: Any*) without spreading it (: _*), so the whole array was
treated as a single template argument instead of being expanded into the
individual placeholders.

For example:

markerLog.info(LogMarker.Security, "{} {} {}", Array("a", "b", "c"))

	base LoggingAdapter	MarkerLoggingAdapter (before)
Output	a b c	ArraySeq(a, b, c) {} {}

The base LoggingAdapter.format1 expands arrays correctly (it calls
formatImpl(t, a.toSeq) directly). The marker copy exists for binary
compatibility reasons (it cannot reach the base trait's private formatImpl),
and when it was written the varargs spread (: _*) was forgotten. The scaladoc
promising that an Array argument "will be expanded into replacement
arguments" was therefore not honored by the marker adapter.

Fixes #3257.

Modification

Spread the array elements as varargs in both Array branches of
MarkerLoggingAdapter.format1, so each element fills its own placeholder, and
documented the reason so it is not regressed:

case a: Array[?] if !a.getClass.getComponentType.isPrimitive => format(t, a.toIndexedSeq: _*)
case a: Array[?]                                             => format(t, a.toIndexedSeq.asInstanceOf[IndexedSeq[AnyRef]]: _*)
case x                                                       => format(t, x)

DiagnosticMarkerBusLoggingAdapter inherits this method and is fixed by the
same change.

Result

• Array arguments to the single-argument marker log template methods are now
  expanded into separate placeholders, matching LoggingAdapter and the
  documented behavior (e.g. "a b c" instead of "ArraySeq(a, b, c) {} {}").
• format1 is private and only its body changed — no public API or
  binary-compatibility change (actor / mimaReportBinaryIssues is clean).

Tests

• sbt "actor-tests/testOnly org.apache.pekko.event.MarkerLoggingSpec" →
  all passed.
• Added directional tests for a non-primitive Array[String] and a primitive
  Array[Int] (the two format1 branches). Both fail before the fix
  (rendering "ArraySeq(a, b, c) {} {}" / "ArraySeq(1, 2, 3) {} {}") and pass
  after.
• sbt "actor/mimaReportBinaryIssues" → no binary issues.

References

Fixes #3257

---

This is an original contribution to Apache Pekko, made under the Apache License
2.0 (i.e. the changes are now Apache licensed). No third-party or Akka-derived
code is included.

[pjfanning]

Is there any chance that since the existing behaviour has been like that for a long time, that we just accept that that is the behaviour? Where in the docs does it say that this change is what users should expect?

[He-Pin]

The problem is this ,it change the default behavior.

[He-Pin]

Not sure if it's safe to change ,let's defer this.

[pjfanning]

Actually, the javadoc says

 /**
   * Message template with 1 replacement argument.
   * The marker argument can be picked up by various logging frameworks such as slf4j to mark this log statement as "special".
   *
   * If `arg1` is an `Array` it will be expanded into replacement arguments, which is useful when
   * there are more than four arguments.
   * @see [[LoggingAdapter]]
   */
  def error(marker: LogMarker, cause: Throwable, template: String, arg1: Any): Unit =
    if (isErrorEnabled(marker)) bus.publish(Error(cause, logSource, logClass, format1(template, arg1), mdc, marker))

So your change is probably correct.

[pjfanning]

lgtm