if possible with nim=>C broken down in different stages (parsing, semantic analysis, C code generation etc)
measuring would really help with applications where compilation speed is critical (eg inim, https://github.com/AndreiRegiani/INim), to see where time is spent in practice, so that optimizations can be applied more effectively.
eg, for an application like inim where the file to comile is typically tiny (ie, each successive line written in interpretter), I'm suspecting that c++ modules that I suggested using in https://forum.nim-lang.org/t/3726#23214 could make a real difference, and make inim more ipython like in terms of speed
There are also other things that have been suggested, eg: