Matthew Reddington created GEODE-7692:
-----------------------------------------
Summary: Sort PdxTypeToTypeIdMap by hash.
Key: GEODE-7692
URL: https://issues.apache.org/jira/browse/GEODE-7692
Project: Geode
Issue Type: Bug
Components: native client
Reporter: Matthew Reddington
PdxTypeToTypeIdMap sorts its keys by "class name", which is not unique and is
the root cause of all our Pdx troubles. The key is a PdxType, and the sorting
order should be determined by a hash of the class name and PdxTFieldTypes. Use
boost::hash_combine to hash the fields and then hash the field hashes.
* This hashing strategy is called a Merkle Tree.
* hash_combine relies on a magic number where each bit has a random chance of
being 0 or 1. This has the effect of spreading consecutive values far apart.
Their random number is the multiplicative inverse of phi - imaginary numbers
are good for this, according to my research.
You can't hash the type ID because this is always 0 for local types, and it's
also 0 for remote types for which we haven't yet received the type id from the
server, or we have constructed a PdxType to perform a lookup to find the type
id in the PdxTypeRegistry, which uses this very map for type id lookup.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
