[LOG4J2-1913] traceExit allocates memory for the input parameter even if it doesn't print it

• Type: Improvement
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Affects Version/s: 2.8.2
• Component/s: API
• Labels: performance
• Flags: Important
• Assignee: Unassigned
• Reporter: Corneliu C
• Created: 18/May/17 14:30
• Updated: 18/May/17 22:45

Description

When calling traceExit(object instance) the underlying code allocates a Message for the input object instance although the log level is not TRACE and there won't be any printing in the end.
With trivial objects is not a problem but using large lists or XML objects results in large memory allocation for nothing which goes to GC and performance impact.

Execution flow is like this:

    @Override
    public <R> R traceExit(final R result) {
        return exit(FQCN, null, result);
    }

next exit calls:

    protected <R> R exit(final String fqcn, final R result) {
        logIfEnabled(fqcn, Level.TRACE, EXIT_MARKER, exitMsg(null, result), null);
        return result;
    }

which calls exitMsg with the input parameter result:

    protected Message exitMsg(final String format, final Object result) {
        if (result == null) {
            if (format == null) {
                return messageFactory.newMessage("Exit");
            }
            return messageFactory.newMessage("Exit: " + format);
        }
        if (format == null) {
            return messageFactory.newMessage("Exit with(" + result + ')');
        }
        return messageFactory.newMessage("Exit: " + format, result);

    }

As it can be seen the exitMsg() method is called and only later the logIfEnabled() is executed.
I assume a log level validation can be applied upfront so as exitMsg is not called for nothing.

Note: this is not the case for other flavors of traceExit which check the log level upfront.

Thank you