SyslogAppender packet splitting for large chunk counts

[jmestwa-coder]

Summary

Fix SyslogAppender packet splitting when large chunk counts cause the (x/y) suffix to exceed the reserved suffix space.

What changed

• Replaced iterator-based splitting with bounded size_t index arithmetic.
• Added suffix-aware chunk sizing to keep emitted payloads within MaxMessageLength.
• Added explicit handling for cases where the full (x/y) suffix cannot fit alongside payload data.
• Replaced fixed apr_snprintf suffix formatting with StringHelper::toString.
• Added defensive guards around packet reservation for extremely large message counts.

Behavior

• Normal packet splitting behavior is preserved for standard message sizes.
• When the suffix cannot fit within MaxMessageLength, the appender now omits the suffix instead of emitting oversized packets.
• A warning is logged when suffix omission occurs.

Tests

Added regression coverage for:

• one-byte remainder splitting boundaries
• large chunk-count suffix growth
• payload length invariant enforcement
• impossible suffix-fit scenarios

Result

Ensures emitted SyslogAppender payloads remain bounded by MaxMessageLength across large message and packet-count edge cases.

[rm5248]

how would this block ever get activated?

[jmestwa-coder]

UNIT_TEST is defined in syslogappendertestcase.cpp, so this path is only enabled for those tests.

The goal was to make the packet-size assertions deterministic without depending on UDP/network behavior. If there’s a preferred way to structure this kind of test in log4cxx, I’m happy to adjust it.

[rm5248]

It would never get compiled with UNIT_TEST defined though, you would have to compile the library with that macro defined.

Instead, it would make more sense to make a helper/utility function that can take a string of arbitrary length and split it up into smaller strings of a given max length. This function would then get called by the SyslogAppender to split the log message up. This function can then be properly tested on its own.

[jmestwa-coder]

Thanks for the suggestion .

I reworked the patch to move the splitting logic into a dedicated helper that SyslogAppender calls directly, removed the UNIT_TEST interception path, and updated the tests to validate the helper behavior directly instead of capturing network writes.

If there’s still a simpler structure you’d prefer here, I’m happy to adjust it further.