bisgardo / go-natural Goto Github PK

There are two small optimization opportunities for Natural:

• There's no need to check if the byte on right hand side is a number if the one on the left hand side isn't.

• Iterate over both numbers simultaneously: If one number is shorter, we only need to keep reading the other number until it's been determined that it's larger.

Add benchmarks to measure if it makes a difference before implementing this.

Computing the numbers contained in the compared strings actually isn't necessary:

Longer numbers will always be larger. Therefore, after skipping leading zeros and resolving the lenghts of the numbers, we can actually just compare the individual digits as regular bytes.

The rule that numbers are always greater than non-numbers should take care of the rest.

Strings are currently indexed on the byte level. As Natural should be correct given general UTF-8 encoded strings, this means that characters may be accessed "within" their multi-byte encoding.

In a UTF-8 encoded string, all bytes with a leading 0 are single-byte (ASCII) characters. This means that all bytes of multi-byte characters start with 1. This means that there cannot be any interference with numbers and multi-byte characters.

The only potential "issue" is that multi-byte characters may not compare to each other the "standard" way: As the characters are compared byte for byte the order will depend on the precise encoding (of which there may be more than one for some characters) rather than comparing the code point.

This is deemed an acceptable shortcoming: As long as the string is normalized (a reasonable requirement), the result is consistent ("longer" characters are "greater").

As a final note, the library is not intended to support character sets that UTF-8 are not compatible with.

Based on the above, the task is the following two items:

1. Verify that all it is true by adding tests that explore the corner cases as well as possible.
2. Add relevant reasoning, preconditions, and "nonstandard" behavior to the documentation.