Can't run specs, compilation fails

[dkhofer]

Hi,

I thought I'd try upgrading some code to Crystal 0.16.0, but the compiler failed with a problem in the matrix code (on which my code relies). I tried running the specs for the matrix shard and the compiler doesn't like what it sees:

$ crystal spec
Error in ./spec/matrix_spec.cr:7: instantiating 'Matrix(T):Class#rows(Array(Array(Int32)))'

      m = Matrix.rows([[1, 2], [3, 4], [5, 6]])
                 ^~~~

in ./src/matrix.cr:60: instantiating 'Matrix(T):Class#new(Int32, Int32)'

    Matrix.new(rows.size.to_i32, rows.first.size.to_i32) do |i, r, c|
           ^~~

in ./src/matrix.cr:38: instantiating 'Matrix(Int32):Class#new(Int32, Int32)'

    matrix = Matrix(T).new(rows, columns)
                       ^~~

instantiating 'Matrix(Int32)#initialize(Int32, Int32)'

in ./src/matrix.cr:28: instantiating 'size()'

    @buffer = Pointer(T).malloc(size)
                                ^~~~

in ./src/matrix.cr:454: undefined method '*' for Int+ (compile-time type is (Int+ | Int32))

    @rows * @columns
          ^


$ crystal --version
Crystal 0.16.0 (Sun May  8 12:21:15 UTC 2016)

I almost think this is a problem with Crystal itself, given that I'm not really sure what an Int+ is, but thought I'd check here first. If you think I should file a bug in the main Crystal repo let me know, I'm happy to do that. Thanks!

[JacobUb]

Yeah thanks for reporting. It's fixed now.
I think the error happened because the type restriction of @rows and @columns was limited to Int which doesn't define a #* method itself although (AFAIK) all its subclasses do.
Int+ means Int or any of its subclasses. According to the docs it would work if Int was an abstract struct (which I thought it was).
Also, the fact that Crystal makes an union of Int+ | Int32 strikes me as oddly redundant but I don't know if it's unintended.

[dkhofer]

Looks like Int is just a regular struct, not abstract: https://github.com/crystal-lang/crystal/blob/master/src/int.cr#L58

Anyway, thanks for the fix!