Show method table for foo( + tab at repl

[simonster]

julia> print(\t
print(io::IO,b::BigFloat) at mpfr.jl:709                      print(io::IO,s::UTF8String) at utf8.jl:138
print(io::IOBuffer,s::SubString{T<:String}) at string.jl:617  print(io::IO,v::VersionNumber) at version.jl:42
print(io::IO,c::Char) at char.jl:56                           print(io::IO,X::AbstractArray{T,N}) at show.jl:895
print(io::IO,ip::IPv4) at socket.jl:30                        print(io::IO,x) at string.jl:3
print(io::IO,ip::IPv6) at socket.jl:85                        print(io::IO,x::Float16) at grisu.jl:119
print(io::IO,n::Unsigned) at show.jl:92                       print(io::IO,x::Float32) at grisu.jl:118
print(io::IO,s::ASCIIString) at ascii.jl:88                   print(io::IO,xs...) at string.jl:4
print(io::IO,s::RopeString) at string.jl:765                  print{mime}(io::IO,::MIME{mime}) at multimedia.jl:17
print(io::IO,s::String) at string.jl:66                       print(xs...) at string.jl:7
print(io::IO,s::Symbol) at show.jl:4

This supersedes #4357.

[simonster]

I just noticed a small issue here: foo(Arra\t replaces the whole string with Array, instead of only the part after the parenthesis. I will fix this.

[stevengj]

Sounds good, once the kinks are worked out!

I should call show_method_table in IJulia too, instead of just sprint(show, methods(o)).

[JeffBezanson]

Really cool!

[StefanKarpinski]

Yes, definitely a nice feature. It would be great to factor out a lot of the common functionality in the REPL and IJulia – and of course replace readline with @loladiro's REPL.jl.

[simonster]

This should work now. I also made it so the readline's built-in filename completion doesn't get called for things like [ that can't be completed, which I'm assuming wasn't intentional.

It sure would be nice if all of this C completion code could be replaced with REPLCompletions.jl, although readline doesn't seem to support specifying what subset of a string should be replaced except via word break characters, which isn't ideal. If REPL.jl is close on the horizon, maybe this is enough for now.

[ivarne]

Nice improvement. I tested it on OSX and it works fine.

[StefanKarpinski]

This does weird stuff for me when I hit tab multiple times.

[simonster]

Can you post what you're typing and what you're seeing? I'm not getting any weirdness.

[StefanKarpinski]

rand(\t then it lists a bunch of stuff and it prints part of the methods list and leaves the cursor at the end. If I just keep hitting \t it leaves me at different random points, and sometimes at the next prompt with no text, sometimes with rand( at the beginning like I would expect. Here are some sequential screen shots where all I've done in between is press \t:

[StefanKarpinski]

[simonster]

@StefanKarpinski Hmm, I can get this to happen, but only on OS X. I can also get it to happen on master with long completion lists (e.g. i\t). I can't get it to happen when I ssh into loopback. I'm not sure what's going on.

[StefanKarpinski]

Honestly, I'm pretty ok with merging and then figuring it out because this is so useful even a little bit broken.

[JeffBezanson]

I bet that's related to #3652. The way we use readline leaves it unable to handle large completions correctly.

[StefanKarpinski]

I would be inclined to say that we should just merge this and wait for a re-implemented repl.

[ivarne]

The implementation does not show methods of operators (like *, /, + and -). There is also an issue with the Unicode operators that are included, and those that are planned. Is there any possibility to access a list of possible operators programmatically, or is that buried inside the flisp parser?

[ivarne]

Why do you both print buf, and return it?

[simonster]

buf is an IO, so that adds a newline to the end of buf without printing anything to stdout.

[ivarne]

Thanks

[ivarne]

I have made a PR at simonster#1 to fix the operator issue.

[stevengj]

Note that tab-tooltip doesn't work for operators in IJulia, apparently because IPython doesn't recognize +( as a function call. cc: @Carreau

[Carreau]

Ah, yes probably. But this is one of the things we need to address in a more deeper way.

Right now, the tooltip is populated with the response to object_info_request which as a parameter take a object to lookup information about. Hence all the processing happend in the frontend, and is brittle as it is based on regexp to both :

• select wether it should complete or prompt tooltip.
• extract the token on the line to request info about it.

I can add +-±... to the list of char than can be part of tokens, but then you might get issues like bar+foo(<tab> that will try to get info on bar+foo instead of foo if there is no space around the +.

We'll probably move toward object_info_request sending the all line (the all cell ?) to the kernel where each language can inspect and decide what to do.

Especially I suppose that you guys, might want tooltip on function(arg, <tab> that lookup the docs only for the signature that match the type of arg for example.

[stevengj]

Yes, sending the whole cell to the front-end, along with the cursor position, would be the most flexible thing. Until that happens, I wouldn't suggest that IPython worry about tooltips for + etc.

[wlbksy]

@JeffBezanson I have problem when compiling julia after this commit, I'm using Mingw64 on Win7

[JeffBezanson]

Ok I pushed a possible fix. Could you try again?

[wlbksy]

Thanks for the quick response. It's fixed now.