This allows for numbers within 18 digits of precision (the vast
majority of numbers) to be represented by just 2 i64s rather than 2 heap
allocated BigInts. In the future these optimizations may become more
aggressive and granular, but for now `grass` is already more than an order of
magnitude faster than `dart-sass`.
prior to this commit, error messages that occurred within idents
beginning with interpolation would be swallowed by error messages for
hex (largely, "Expected identifier.")
Before this commit, escape sequences above std::char::MAX ('\u{10ffff}')
would overflow and cause a panic. This commit replaces an `unwrap` with
`ok_or` and a clearer error message. This message will likely change
in the future in order to better conform to the `dart-sass` implementation
which currently also fails to cleanly handle this overflow.
See https://github.com/kaj/rsass/pull/73
This release contains significant (>10x) improvements for WASM speed.
Performance is now comparable to libsass bindings with `node-sass` as
well as `dart-sass` with dart2js. It is, however, roughly 4x slower than
native `grass`.
grass differs from rsass in that all numbers are bigints. this makes our
fomatting of numbers 6x slower than rsass and barely faster than
node-sass. ideally in the future, we would just use plain float
formatting. i've learned that dart-sass no longer configures precision as a cli
flag, so there is potentially some manual loop unrolling or
optimizations we could make knowing that all numbers must be a max of 10
digits. the next release of num_rational should contain an
implementation of to_f64() for BigRational, so that may solve our
problems.
