zio.json.internal

A Reader that can retract and replay the last char that it read.

This is essential when parsing contents that do not have a terminator character, e.g. numbers, whilst preserving the non-significant character for further processing.

Total, fast, number parsing.

The Java and Scala standard libraries throw exceptions when we attempt to parse an invalid number. Unfortunately, exceptions are very expensive, and untrusted data can be maliciously constructed to DOS a server.

This suite of functions mitigates against such attacks by building up the numbers one character at a time, which has been shown through extensive benchmarking to be orders of magnitude faster than exception-throwing stdlib parsers, for valid and invalid inputs. This approach, proposed by alexknvl, was also benchmarked against regexp-based pre-validation.

Note that although the behaviour is identical to the Java stdlib when given the canonical form of a primitive (i.e. the .toString) of a number there may be differences in behaviour for non-canonical forms. e.g. the Java stdlib may reject "1.0" when parsed as an BigInteger but we may parse it as a 1, although "1.1" would be rejected. Parsing of BigDecimal preserves the trailing zeros on the right but not on the left, e.g. "000.00001000" will be "1.000e-5", which is useful in cases where the trailing zeros denote measurement accuracy.

BigInteger, BigDecimal, Float and Double have a configurable bit limit on the size of the significand, to avoid OOM style attacks, which is 128 bits by default.

Results are contained in a specialisation of Option that avoids boxing.