David: Thanks for the help! I'm not sure how to implement your suggestions, however.
My main reason for using eval() is to convert from a Symbol to a Type. If they supplied a full expression like (1 + 1) as the default value for an argument, well, Julia supports that anyway, so eval()ing it once inside the macro and saving the result for all future calls matches the behavior of python (I don't know if Julia matches python behavior here). I'm really not sure what to make of your statement "the unfortunate need to throw in esc in a couple places to trick Julia into using the intended context for ptype and default." The macro seems to be working well right now; where do I need to add esc()? On Friday, May 9, 2014 9:10:35 AM UTC-4, David Moon wrote: > > On Thursday, May 8, 2014 12:47:03 PM UTC-4, Adam Smith wrote: >> >> .... >> Then I was able to develop a flexible macro that better supports the >> standard syntax, and I think is rather elegant: >> # Flag a function argument of any type as optional by generating a Union. >> # If a default value is not defined, it assumes "nothing". >> # Usage: >> # function(x::Int, @maybe y::MyType) >> # function(x::Int, @maybe y::MyType=someval) >> macro maybe(argexpr) >> default = nothing >> if argexpr.head == :(=) >> argexpr, default = argexpr.args >> default = eval(default) >> end >> @assert argexpr.head == :(::) >> name, ptype = argexpr.args >> deftype = typeof(default) >> Expr(:kw, :($name::Union($deftype, $ptype)), default) >> end >> > > If you're going to eval the default expression, then you need to quote it > in the Expr you return so it doesn't get evaluated again at run time. > Except for that, and the unfortunate need to throw in esc in a couple > places to trick Julia into using the intended context for ptype and > default, this looks great. > > It would be much cleaner not to evaluate the default expression, just plug > it into the returned Expr and let it get evaluated at run time. But then > instead of calling typeof, you would need to call into compile-time type > inference to get the deftype. I am too lazy to read through all the > documentation to try to figure out if there is an API that > exposes compile-time type inference. But if people are going to write > macros like this, there should be. > > I recommend updating the gist after you have a good fix for that. > > --Dave Moon >
