matthieu-vergne / benchmark Goto Github PK

For CsvStoring, the full file content is stored in memory and processed in order to look how many lines have already been written (when starting the benchmark). It would be surely far better to read and count on the fly, without storing anything, especially for big files, which take a lot of time to proceed.

Now that this lib has been used in some contexts, a complete revision should be made to simplify its use. In particular, several perspectives should be considered:

1. task creation: inputs, outputs, code
2. task linking: relate output to input(s)
3. internal use (code): get inputs, set outputs
4. external use (link): get output, set input(s), transfer upon output setting or task termination
5. task composition (ultimately, a benchmark is a composed task)
6. task running

The point is to have an architecture which allows to be concise in each perspective, access only the relevant methods depending on the perspective, exploit generics to retrieve types easily, etc.

The use of factories and builders is a must, and progress monitoring is recomended.

As we have the possibility to define a RESET input, it would be nice to set an EXECUTION_TIME output which considers the full time of the execute() method.