Thank you for your answer. I knew your package and I pay great respect for the
incredible work you've done (and for the one _mratsim_ has done with
`arraymancer`). Since `datamancer` codebase is quite large and use macros and
some other more advanced / performance optimized stuff (buf reading), I
couldn't get if my "toy" example was completely off track or was at least
catching the main points about this topic (data type inference and assignment
at runtime). I agree, CSV is a poor data exchange format... but incredibly
widespread, because convenience and simplicity often is more succesful than
strictness (the same may be applied to dynamic typing vs strict static typing).
Anyway Nim compiled code is so blazing fast that sometimes also higly
un-optimized code is fast enough when compared with alternatives.
