peco not installed

[ghost]

I tried:

] add InteractiveCodeSearch
using InteractiveCodeSearch
@search show

I get the following error:

┌ Warning: Matcher peco not installed.
│ See https://github.com/peco/peco for how to install peco.
│
└ @ InteractiveCodeSearch C:\Users\phg\.julia\packages\InteractiveCodeSearch\bvRfv\src\InteractiveCodeSearch.jl:559
ERROR: IOError: could not spawn `peco`: no such file or directory (ENOENT)

How do I install peco? Is there a way for peco to be installed automatically together with the package?

[tkf]

How do I install peco?

As mentioned in https://github.com/peco/peco#installation, just download it from https://github.com/peco/peco/releases and place it to somewhere in your PATH.

Is there a way for peco to be installed automatically together with the package?

It is possible, maybe by using https://github.com/JuliaPackaging/BinaryProvider.jl. I'm not planning to do this right now but if you or somebody get this work I'd be happy to merge the PR.

[rapus95]

Ran into the same ^^ Since the Artifact system arrived, this might be even easier. :D I might eventually give it a shot if I find time 😅

[tkf]

Thanks, that would be great!

[rapus95]

I guess you want to keep Julia LTS compat, thus, even if I can make an artifacts solution, I have now clue how to backport that for pre 1.3 🤔

[tkf]

I don't know much about the Artifacts system but maybe something like this would work?

@static if VERSION < v"1.3"
    const peco_artifact = nothing
else
    using Pkg.Artifacts
    const peco_artifact = artifact"peco"
end

I think it's OK to throw a warning/error in old julias.

[rapus95]

Btw, when searching, where to put that code to, I found that part:

@static if VERSION < v"0.7-"
[...]

You might want to consider dropping all that pre 1.0 stuff similar to Compat's decision https://github.com/JuliaLang/Compat.jl/pull/665

[tkf]

Unlike Compat.jl there hasn't been much update in InteractiveCodeSearch.jl after julia 1.0 was out. So that's why it still supports 0.6. But yes, it's better be dropped now.

[rapus95]

Finally after just too much time for that amount of code 😄

https://github.com/tkf/InteractiveCodeSearch.jl/pull/10

Though, on julia pre 1.3 the package is somewhat unnecessary (but shouldn't fail is its not used)... But I don't know of conditional packages

[tkf]

Regarding https://github.com/JuliaPackaging/Yggdrasil/pull/377, do we need to keep updating P/peco/build_tarballs.jl to track peco releases? Maybe it's OK. I'm just wondering how it'd work.

[tkf]

So, from the discussion in slack, it looks like we need to design an interface to make CONFIG.interactive_matcher more generic.

Option 1: (::IO) -> ::String

A straightforward option is to just factor out current use of run_matcher:

# Internal function to be used from `_choose_method`:
_run_matcher(input) = run_matcher(CONFIG.interactive_matcher, input)

# Let's still allow `CONFIG.interactive_matcher :: Cmd`
function run_matcher(matcher::Base.AbstractCmd, input::IO)
    maybe_warn_matcher(matcher)
    return String(read_stdout(input, matcher))
end

# Otherwise assume that it's a callable/function
run_matcher(matcher, input::IO) = matcher(input)

Option 2: (::Iterator{String}) -> ::String

Maybe option 1 is too low-level? I think a nicer abstraction is to pass an iterator of strings.

# `strings` is an iterable
_run_matcher(strings) = run_matcher(CONFIG.interactive_matcher, strings)

function run_matcher(matcher::Base.AbstractCmd, strings)
    maybe_warn_matcher(matcher)
    return read_stdout(matcher) do io
        for line in strings
            println(io, line)
        end
    end
end

run_matcher(matcher, strings) = matcher(strings)

(This creates a lot of intermediate Strings. But if we want performance afterwards, we can always create an AbstractString subtype that is non-allocating and overloads print(::IO, ::CustomStringType).)

In both cases, I think we can use peco_jll via:

@static if VERSION < v"1.3"
    import peco_jll

    peco(input) = peco_jll.peco() do cmd
        run_matcher(cmd, input)
    end
else
    peco(input) = run_matcher(`peco`, input)
end

I'm inclined to Option 2. What do you think? Is there a better way to do this?

[dpo]

Have a look at how Ipopt.jl does it. They have a build.jl that installs the binaries for Julia < 1.3 and relies on the artifact system otherwise. It would be great if peco were installed automagically.