Performance comparison with function composition

[Oblynx]

@chain rewrites each expression to assign to temporary variables:

Chain.jl/src/Chain.jl

Line 46 in 747639e

new_expr = Expr(Symbol("="), replacement, new_expr)

Can this lead to a performance tradeoff compared with plain-Julia function composition?
Especially in the case of broadcasting, I suppose this should break syntactic loop fusion, or not?

[jkrumbiegel]

I think it can, yes, but the things I made it for (Dataframe manipulation mostly) don't work with fused expressions anyway. That's because you can't express a groupby combine strategy as a kernel function iterating once over your original dataframe.

So yes, don't use this as a replacement for broadcasting syntax. To me it's also important that error messages are informative, which works better with the stepwise approach, so it's just a tradeoff

[Oblynx]

I totally agree and thanks a lot for your package 😊

I've been using lazy.jl @> so far, but I'm gonna give this a try because of the debugging syntax. I just wanted to clear out the inherent limitation.