A Performance Question

[gmseed]

Hi

I recently stumbled on EvalEx and really impressed by both the library and documentation.

I have a data-processing application and EvalEx works well but I had a performance question.

In most real-world applications Java's type double will more than suffice and was wondering if going straight to double rather than first to BigDecimal and then to double would offer any performance improvement?

Or, is the performance hit when evaluate() is called as opposed to doubleValue() from BigDecimal.

In data-processing or time-series applications even bit of performance is key, and the reason for my post.

Thanks
Graham

[uklimaschewski]

Hi Graham,

thanks for your feedback and good question.

EvalEx is designed to be precise and safe, so it uses BigDecimal as the internal number type. This avoids rounding errors and makes the library reliable also for financial or other exact use cases. In many business and regulatory environments, exact decimal results are required, and double would not be accepted.

Converting the result with doubleValue() is very fast compared to the evaluation itself. Most of the time is spent in parsing and evaluating the expression, not in the final conversion.

If your application only needs double and very high performance, EvalEx may not be the fastest choice, because BigDecimal is slower than primitive doubles. But it gives you correctness and flexibility.

[gmseed]

Thanks for your reply.

I came across the embedded compiler, Janino, and added support for this library alongside EvalEx. Janino's expression evaluator can use any supplied function but naturally exploits Java's Math class, with most methods supporting type double. And not having to use BigDecimal and compiling to byte code it is fast.