eclipse-aascw / aas-core-codegen / 35359897208

90%

push

github

Test the XSD edge cases of the primitive types (#728)

We add tests to pin down compatibility among the SDKs and their precise
shared behavior.

A recorded example under `Xml/Expected` has to serialize back to itself.
That rules out every interesting double, as
the six targets do not agree on how to write one -- the largest double
goes out as `1.7976931348623157E308` from Java and as
`1.7976931348623157e+308` from the others, and even 100.0 is `100` in
TypeScript -- and it rules out a value which is read in one form and
written in another: `1` for a boolean comes back as `true`, and
a base64 with whitespace between its characters comes back without it.
Those are asserted in code instead, by the generated tests: each target
reads the first recorded example with the content of one element
replaced, and asserts the *value*, never the text. That covers `1e400`
and `-1e400` rounding to an infinity, `1e-400` rounding to zero,
the three named literals, the four literals of `xs:boolean`, and
a `xs:base64Binary` both with whitespace inside it and empty.

We cover with additional test data the spellings which are not
`INF`, `-INF` or `NaN`; numerals which are no numerals, such as `0x10`,
`1.0abc` and `1.0e`; an internal space, which survives the collapsing as
a single space and leaves `2  3` no number; both ends of the `xs:long`
range; the four literals of `xs:boolean` against `TRUE`, `True`, `yes`
and `2`; and one per rule of the `xs:base64Binary` production -- a length
which is no multiple of four, a character outside the alphabet, an equals
sign in the middle, and a `SG==` whose dropped bits are not zero.

XSD 1.1 accepts `+INF`, while XSD 1.0 rejects it. For the sake of
robustness, we also accept `+INF`.

One thing could not be covered, and is left alone deliberately.
A non-finite JSON literal can not be a fixture at all, as every target
parses the file as JSON before handing it over, so `NaN` breaks
the harness instead of reaching the reader.

222 of 222 new or added lines in 8 files covered. (100.0%)

34975 of 40722 relevant lines covered (85.89%)

2.58 hits per line