ArgumentError when decoding an int > 134_217_727

[beam-community]

ArgumentError when decoding an int > 134_217_727

[tr00gle]

Disregard Question 3: I commented out the @spec and @spec_match and ran the tests as is in the source: they work.

Then, I changed 5_000_000 to 500_000_000 and received the argument error again:

  1) test decode (primitive) integer (AvroEx.Decode.Test)
     test/decode_test.exs:22
     ** (ArgumentError) argument error
     code: assert {:ok, 500_000_000} = @test_module.decode(schema, big)
     stacktrace:
       (avro_ex 1.1.0) lib/avro_ex/decode.ex:238: AvroEx.Decode.variable_integer_decode/3
       (avro_ex 1.1.0) lib/avro_ex/decode.ex:64: AvroEx.Decode.do_decode/3
       (avro_ex 1.1.0) lib/avro_ex/decode.ex:12: AvroEx.Decode.decode/2
       test/decode_test.exs:33: (test)

[tr00gle]

Ok, so after a bunch more research, it seems that the threshold is 134_217_727.

134_217_727 passes, but 134_217_728 fails. What are the constraints here that it's written this way? Would it be possible to catch these argument errors and treat them as type long automatically? I'm not sure what the potential ramifications are here, but I'm happy to research/do whatever I can to assist.

[tr00gle]

After more research, and studying the avro-official decoding/encoding implementations in ruby and javascript, and it seems like the implemented logic in this library is pretty similar, except swapping while loops for recursive calls. A key difference is that the ruby implementation is that the integer-related logic delegates to the same logic for type long, and I wonder if that might be helpful here?

Also, the inability to decode integers that have an absolute value larger than 134_217_727 seems like an unexpectedly low threshold still. A few more follow up questions:

1. Since tests tend serve as a form of documentation, would it be appropriate to update the test cases for decoding integers to use the value 134_217_727 instead of 5_000_000? I have a branch and commit ready to go, but if it's not something we want, I won't open a PR.
2. Should we update the tests for type long accordingly as well?
3. Again, these numbers aren't particularly large. Has nobody else run into these issues either?

[tr00gle]

NICE!