Crystal: Error in Type cast from JSON Array

[raydf]

test = nil
test = JSON.parse %[
  {
    "test": {
      "test_array": [
        {
          "test_hash": 1
        }
      ]
    }
  }
]

# pp JSON::Any.new(test.try &.["test"]?.try &.as_h?.try &.["test_array"]?).try &.as_a?
pp test.try &.["test"]?.try &.as_h?.try &.["test_array"]?.try &.as?(Array(JSON::Type))

Workaround in the commented line with JSON::Any.new()

Invalid memory access (signal 11) at address 0x7fff00000006
[4521026555] *CallStack::print_backtrace:Int32 +107
[4521010759] __crystal_sigfault_handler +55
[140735671387434] _sigtramp +26
[4521182122] *Pointer(Array(JSON::Type) | Bool | Float64 | Hash(String, JSON::Type) | Int64 | String | Nil)@Pointer(T)#[]<Int32>:(Array(JSON::Type) | Bool | Float64 | Hash(String, JSON::Type) | Int64 | String | Nil) +10
[4521181823] *Array(JSON::Type)@Array(T)#unsafe_at<Int32>:(Array(JSON::Type) | Bool | Float64 | Hash(String, JSON::Type) | Int64 | String | Nil) +31
[4521181221] *Array(JSON::Type)@Array(T)#to_s<String::Builder>:(String::Builder | Nil) +357
[4521180857] *Array(JSON::Type)@Array(T)#inspect<String::Builder>:(String::Builder | Nil) +9
[4521180818] *Array(JSON::Type)@Object#inspect:String +66
[4520976408] __crystal_main +2088
[4521010504] main +40

Env:

MacOs X
Crystal 0.19.2 (2016-09-16)

[oprypin]

No idea what's going on here or how you managed to get this error (does not happen for me).

Here is an easier way to deal with reading the JSON structure:

test["test"]["test_array"].as_a

[RX14]

Sorry, I cannot reproduce your bug either. Does that _exact_ code snippet reproduce the bug for you?

[raydf]

@RX14, @BlaXpirit, please see the environment part of the "issue". You are not using the new release of crystal:
https://play.crystal-lang.org/#/r/1anv
https://play.crystal-lang.org/#/r/1anu

[raydf]

@BlaXpirit , in comments (#) of the "issue" code there's a parsing similar to your code, but with the safety of nil checking, and i stated it like a workaround.
Still there's an error in the type casting code in the last release that worked in previous versions.

[RX14]

@raydf I can assure you I was using 0.19.2 when I tested that snippet and it ran fine.

[RX14]

Segfaulting only on specific OSes and architectures points to a codegen issue. I don't have a mac to see if this happens on all macs.

[raydf]

Thanks @RX14, @BlaXpirit for reporting that in linux 64bit is working correctly. I'll keep using the workaround.

[asterite]

I was able to reproduce it, and found related bugs. I'll probably debug and fix these next week.

[asterite]

Reduced:

class Foo
end

x = Foo.new.as(Int32 | Foo)
x.try &.as(Foo)

It gives a codegen error in this case, but it's essentially the same bug (fixing the above fixes this)

[raydf]

Thanks @asterite.