Release 10.7.1.2
Cumulocity IoT Transport

PAM-32554
Breaking change in GenericResponse behavior should be configurable to old behavior.
10.7.0 introduced a breaking change in the contents of the com.apama.cumulocity.GenericResponse.body member. This fix introduces a mechanism to revert to the old behavior. This can be done by setting a property in the correlator on startup: analyticsBuilder.genericResponse.body=payload. This may be needed for compatibility with applications running in a Streaming Analytics application in Cumulocity IoT.
Connectivity and Adapters

PAM-32455
Shutting down a chain containing a queue should not lose events that are on the queue.
If the Batch Accumulator codec has any messages queued to send towards the host, it waits to flush them before shutting down. This is not true if the host itself is shutting down (rather than just the chain).

PAM-32451
Chain::isStartThread threadlocal should be a static to avoid hitting PTHREAD_KEYS_MAX limit.
Prior to this fix, each connectivity chain requested its own thread-local key for chain management purposes. When a large number of chains is created (for example, more than 1000), the maximum number of thread-local keys in the system may be reached, with consequential effects on other functionality in the system wishing to use thread-local memory. On Linux, a call to pthread_key_create() would return EAGAIN. On Windows, a call to TLSAlloc() would return TLS_OUT_OF_INDEXES. The fix ensures that only one thread-local key is requested for the management of all connectivity chains.

PAM-32345
Crash in HTTP concurrent chains on ARM.
On ARMv7 platforms using the HTTP server connectivity plug-in configured for concurrent chains, the correlator dumped core when those chains were started. This is now resolved.
Correlator

PAM-32485
deleteAll does not actually remove coverage information.
A change in 10.7.1 retained coverage information for transient/deleted files, except on deleteAll. A bug caused it to also be retained on deleteAll, but this has now been fixed.

PAM-32417
LLVM includes n^2 copies of the clang version in .comment sections.
Applications running the compiled runtime on Linux can experience high memory usage and long compilation times. When using a compilation cache, then they can also experience very large cache directories. This happens with complex EPL applications with very deep dependency graphs between EPL types. Redundant object file information has been removed, improving the size of cache directories, injection times and memory usage for the compiled runtime.