tests fail, no method isxdigit(Int32)

[mschauer]

While testing a bugfix, I ran into this unrelated error, which lets the test suite fail in test string.jl, line 62

julia> escape_string(string('\u1ffd', 'x'))
"´x"
julia> escape_string(string('\u1ff0','x'))
ERROR: no method isxdigit(Int32)
 in need_full_hex at string.jl:837
 in print_escaped at string.jl:843
 in sprint at io.jl:462
 in escape_string at string.jl:856

[pao]

This test appears to pass on Travis, and isxdigit is exported from Base. Maybe you need to rebuild your system image?

[ihnorton]

please send:

• versioninfo()
• methods(isxdigit)

[mschauer]

julia> methods(isxdigit)
# 2 methods for generic function "isxdigit":
isxdigit(c::Char) at string.jl:835
isxdigit(s::String) at string.jl:836

julia> versioninfo()
Julia Version 0.3.0-prerelease+490
Commit f8f3190 (2013-12-15 07:16 UTC)
Platform Info:
  System: Linux (i686-linux-gnu)
  WORD_SIZE: 32
  BLAS: libopenblas (DYNAMIC_ARCH NO_AFFINITY)
  LAPACK: libopenblas
  LIBM: libopenlibm

[mschauer]

I rebuild everything after git clean -fdx and the problem persists. I also fail at
length(1:4:typemax(Int64))==2305843009213693952 in test-ranges.

Does travis cover 32 bit machines?

[mschauer]

See also #4813

[staticfloat]

I can confirm that this is a 32-bit problem. It happens on 32-bit wine and 32-bit ubuntu builds.

[staticfloat]

This is where I admit that I don't know what's going on:

julia> z = unescape_string(string('\ufff','\x0'))
"Evaluation succeeded, but an error occurred while showing value of type UTF8String:
ERROR: no method display(UTF8String)
 in display at multimedia.jl:158

(Note that this "no method" error is really just the underlying error reported above being caught by an overzealous try/catch)

julia> Base.need_full_hex(z,1)
ERROR: no method isxdigit(Int32)
 in need_full_hex at string.jl:838

julia> Base.need_full_hex(z,4)
ERROR: no method isxdigit(Int32)
 in need_full_hex at string.jl:838

julia> Base.need_full_hex(z,5)
false

However, if we just plug in the definition of Base.need_full_hex, we get:

julia> !done(z,1) && isxdigit(next(z,1)[1])
false

julia> !done(z,4) && isxdigit(next(z,4)[1])
false

julia> !done(z,5) && isxdigit(next(z,5)[1])
false

If we wrap that definition in a function though, we get an error again (instead of the correct value which is indeed false):

julia> g(s,i) = !done(s,i) && isxdigit(next(s,i)[1])
g (generic function with 1 method)

julia> g(z,1)
ERROR: no method isxdigit(Int32)
 in g at none:1

julia> g(z,4)                                                                   
ERROR: no method isxdigit(Int32)
 in g at none:1

julia> g(z,5)                                                                   
false

[StefanKarpinski]

My very vague suspicion is that this might be due to @loladiro's recently merged tuple optimization branch.

[Keno]

I'll grab myself a 32bit system and see what's going on.

[vtjnash]

This seems to be a reappearance of #4066:

julia> g(s,i) = !done(s,i) && isxdigit(next(s,i)[1])
g (generic function with 1 method)

julia> z = unescape_string(string('\ufff','\x0'))
"Evaluation succeeded, but an error occurred while showing value of type UTF8String:
ERROR: no method display(UTF8String)
 in display at multimedia.jl:158

julia> code_lowered(g,typeof((z,1)))
1-element Array{Any,1}:
 :($(Expr(:lambda, {:s,:i}, {{},{{:s,:Any,0},{:i,:Any,0}},{}}, quote  # none, line 1:
        unless !(done(s,i)) goto 0
        return isxdigit(getindex(next(s,i),1))
        0: 
        return false
    end)))

julia> code_typed(g,typeof((z,1)))
1-element Array{Any,1}:
 :($(Expr(:lambda, {:s,:i}, {{},{{:s,UTF8String,0},{:i,Int32,0}},{}}, quote  # none, line 1:
        unless top(box)(Bool,top(not_int)(top(slt_int)(endof(s::UTF8String)::Int32,i::Int32)::Bool))::Bool goto 0
        return isxdigit(tupleref(top(tuple)(getindex(s::UTF8String,i::Int32)::Char,top(box)(Int,top(add_int)(top(box)(Int,top(add_int)(i::Int32,1))::Int32,arrayref(top(utf8_trailing),top(box)(Int,top(add_int)(top(box)($(Int32),top(zext_int)($(Int32),arrayref(top(getfield)(s::UTF8String,:data)::Array{Uint8,1},i::Int32)::Uint8))::Int32,1))::Int32)::Int32))::Int32)::(Char,Int32),1)::Union(Int32,Char))::Bool
        0: 
        return false
    end)))

(in code_llvm, the output of the tupleref gets boxed via jl_box_int32, then passed to jl_apply_generic)

If I turn off inlining, it fails while emitting a reference to the 2nd element of ((Base.DevNullStream, Base.DevNullStream, Base.DevNullStream), Bool, Bool) with the complaint that julia-debug-basic: /home/jameson/julia/usr/include/llvm/IR/Instructions.h:704: llvm::Type* llvm::checkGEPType(llvm::Type*): Assertion Ty && "Invalid GetElementPtrInst indices for type!"' failed.`

[vtjnash]

64-bit test code:

julia> bitstype 64 Int5142

julia> f(x::Int64) = "bad"
f (generic function with 1 methods)

julia> f(x::Int5142) = "good"
f (generic function with 2 methods)

julia> function g(a)
       x=a?(int64(0),reinterpret(Int5142,int64(0))):(int64(1),reinterpret(Int5142,int64(1)))
       f(x[2]::Int5142)
       end
g (generic function with 1 method)

julia> g(true)
"bad"

julia> function h(a)
              x=a?(int64(0),reinterpret(Int5142,int64(0))):(int64(1),reinterpret(Int5142,int64(1)))
              x[2]::Int5142 
              end
h (generic function with 1 method)

julia> h(true)
ERROR: type: h: in typeassert, expected Int5142, got Int64
 in h at none:3