Show with Type{T <: X} messes with Base.show

[SimonDanisch]

Hi,
just run into the following:

abstract Mesh

function Base.show{M <: Mesh}(io::IO, ::Type{M})
end

julia> println(methods(similar))
#42 methods for generic function "similar":
similar(r::Range{T}, T::Type{T}, dims::Tuple{Vararg{Int64}}) at range.jl:252
similar(r::Range{T}, T::Type{T}, dims::Tuple{Vararg{Integer}}) at range.jl:251
similarERROR: type DataType has no field body
 in show at show.jl:76 (repeats 2 times)
 in show_delim_array at show.jl:219
 in show at show.jl:73
 in show_type_parameter at show.jl:82
 in show at show.jl:92
 in show_type_parameter at show.jl:82
 in show at show.jl:92
 in show at expr.jl:56
 in show_type_parameter at show.jl:82
 in show at show.jl:92
 in print at strings/io.jl:5
 in print_to_string at strings/io.jl:19
 in argtype_decl at methodshow.jl:26
 in arg_decl_parts at methodshow.jl:38
 in show at methodshow.jl:44
 in show_method_table at methodshow.jl:69
 in print at strings/io.jl:5
 in print at strings/io.jl:6
 in println at strings/io.jl:8
 in println at stream.jl:242
 in println at strings/io.jl:11

Am I using some forbidden technique, or is this simply a bug?!

versioninfo()
Julia Version 0.4.0-dev+6492
Commit a6c78b6* (2015-08-04 07:48 UTC)
Platform Info:
  System: Windows (x86_64-w64-mingw32)
  CPU: Intel(R) Core(TM) i5-4200U CPU @ 1.60GHz
  WORD_SIZE: 64
  BLAS: libopenblas (USE64BITINT DYNAMIC_ARCH NO_AFFINITY Haswell)
  LAPACK: libopenblas
  LIBM: libopenlibm
  LLVM: libLLVM-3.3

Best,
Simon

[timholy]

It's generally fine to play with show for values, but messing with show for types is frowned upon (speaking as the subject of the frowning).

[SimonDanisch]

I see... Still, there is a difference between frowned uppon and destroying
Base functionality.
So should it be deprecated? I'm not really seing the point though, besides
making show more predictable to the user.

From how I see it, show does not give any guarantees like "this yields
always the same printout for any type", though. So I'm not sure why it
should be different for types.

Why don't we let users overload show and use dump/xdump for the full
description?

The printout of my Mesh type is pretty much unusable without overloading
it.

On 4 Aug 2015 20:24, "Tim Holy" notifications@github.com wrote:

It's generally fine to play with show for values, but messing with show
for types is frowned upon
JuliaMath/FixedPointNumbers.jl#9 (comment)
(speaking as the subject of the frowning).

—
Reply to this email directly or view it on GitHub
#12456 (comment).

[Keno]

If it's legal to define, it should not mess with Base printing, so this is a bug.

[simonster]

In any case it seems like it's probably a bad idea to define because it will cause show to be specialized/compiled for each type-valued argument it may be called on, instead of just being specialized/compiled once for DataType.

[JeffBezanson]

Looks similar to #8163

[vtjnash]

this was fixed along with #8163