SIGNATURES randomly generated

[amontoison]

I tried to use DocStringExtensions.jl with the package Krylov.jl but I encountered some problems. When I build the docs, the signature of only two random Krylov solvers is generated. Do you have an idea where the problem comes from?
JuliaSmoothOptimizers/Krylov.jl#208

[MichaelHatherly]

What kind of output are you getting when using ? mode in the REPL? Are the templates showing up there?

[amontoison]

Yes, but only for one solver. The signature is only given for the first method when using ? after using Krylov.

[exaexa]

Hi! Reviving this issue-- I see the same problem with TYPEDSIGNATURES in PikaParser package here. Missing signatures can be seen e.g. here:
https://lcsb-biocore.github.io/PikaParser.jl/dev/reference/#PikaParser.followed_by-Tuple{G}%20where%20G
and here
https://lcsb-biocore.github.io/PikaParser.jl/dev/reference/#PikaParser.parse_lex-Union{Tuple{I},%20Tuple{T},%20Tuple{G},%20Tuple{PikaParser.Grammar{G,%20T},%20I}}%20where%20{G,%20T,%20I}

Quite surprisingly, a different set of signatures is missing (lex instead of parse_lex) when I build the same docs from the same source locally.

Regarding this:

What kind of output are you getting when using ? mode in the REPL? Are the templates showing up there?

No, the missing signatures are not present even in REPL.

[exaexa]

...turns out methodgroups might be to blame, this is from a quick debug:

┌ Info: in format(::TypedMethodSignatures, ...)
│   binding = PikaParser.some
│   typesig = Tuple{G} where G
│   modname = PikaParser
│   func = some (generic function with 1 method)
└   groups = Vector{Method}[]

🙄

[exaexa]

looks like in some cases the unionall is missing in typesigs in methodgroups, which breaks the in typesigs check:

┌ Info: in methodgroups
│   typesigs =
│    1-element Vector{Any}:
└     Tuple{G}
┌ Info: in methodgroups filter
│   m = some(x::G) where G in PikaParser at /home/exa/work/PikaParser.jl/src/frontend.jl:134
│   ismod = true
│   m.sig = Tuple{typeof(PikaParser.some), G} where G
└   Base.rewrap_unionall(Base.tuple_type_tail(m.sig), m.sig) = Tuple{G} where G
┌ Info: in format(::TypedMethodSignatures, ...)
│   binding = PikaParser.some
│   typesig = Tuple{G} where G
│   modname = PikaParser
│   func = some (generic function with 1 method)
└   groups = Vector{Method}[]

[exaexa]

Ok I probably found the problem, looks like a sneaky broadcast:

julia> sig = (Tuple{G} where G)
Tuple{G} where G

julia> Base.rewrap_unionall.(DocStringExtensions.uniontypes(Base.unwrap_unionall(sig)), sig)
1-element Vector{DataType}:
 Tuple{G}

julia> Base.rewrap_unionall.(DocStringExtensions.uniontypes(Base.unwrap_unionall(sig)), Ref(sig))
1-element Vector{UnionAll}:
 Tuple{G} where G

I'll check how much this breaks all the other stuff and send a patch. :]