Question about performance of Argu

[matthid]

Description

I know that Argu uses reflection and therefore is kind of slow. I just wanted to ask if there are other ways of achieving performance besides doing stuff like fsprojects/Paket@827c282

For paket restore I consider this way too slow for what I'm trying to achieve. Basically the above commit improved performance from around 3secs to about 700ms

Maybe we need a type-provider for argument parsing but we don't have type arguments yet
/cc @dsyme

[eiriktsarpalis]

It would probably make sense to profile performance characteristics of argu on a large CLI interface like paket. Paket does a great deal of reflection-based precomputation at initialization which may not be required in the scope of running a single command. I think startup time could be improved by making certain aspects of union case metadata extraction lazy. I followed this approach when implementing subcommands and could certainly be retrofitted to other aspects of reflection-based extraction.

[matthid]

Ok I might be able to take a look at this thanks for the hints. So you think it will be enough and we don't need a radically different approach?

[eiriktsarpalis]

So, because of the peculiar nature of this library, namely using reflection to parse a single string of arguments, I'm not entirely convinced that reflection might be the dominant factor in Argu performance. I wouldn't be too surprised if runtime loading of the library itself is what takes the longest time. Perhaps comparing 32bit vs. 64bit RyuJIT startup times might reveal something there.

[matthid]

Some progress has been made in #109 and #105