Optimized expressions are not cached

[bjorne]

By default JMESPath#search caches parsed expressions in order to improve performance. However, it also takes care to call #optimize on each expression. The optimized expressions are not cached, meaning that for every #search the optimization takes place.

According to my (limited) tests, caching the optimized expressions increases the performance by 10-30 times.

I see two ways to solve this;

• Cache the optimized expressions within Runtime, instead of caching the non-optimized expressions in the CachingParser, or
• Make #optimize memoize its result, so that sub-sequent calls can return the value directly without the overhead

I'll happily help with the coding, but wanted to get some feedback first.

[bjorne]

Sorry, I misunderstood things a bit. Since JMESPath.search creates a new Runtime for every call, it also creates a new CachingParser for each call. This should mean that expressions are not actually cached, and consequently my estimated performance increase would not be as significant as I stated just from caching the optimized expressions.

[iconara]

I suggest removing the CachingParser completely and instead introduce a CachingRuntime that provides the expression caching and a constant DEFAULT_RUNTIME that is used by JMESPath.search.

[iconara]

It seems like the caching optimization got lost in two steps: first I introduced the optimization step and missed that the result of the optimization wasn't cached, and then 5703438 changed things so that the caching parser is created for each new runtime.