Not critical, but it would be nice to control

• gcc in ubuntu on 64-bit machine uses long of 8 bytes inside of num struct
• mingw with its own gcc seemingly has long of 4 bytes instead.

Ideally it would be great to specify explicitly during the build time.

This may require testing, but it seems we have single variable for backchar (like ungetc), despite having many ports. So backchar called on stdin and then immediately on some file should be owerwritten, which is no good.

There is no efficient way to copy vector (besides (list->vector (vector->list s))) but R7RS introduces vector-copy, vector-copy! and vector-append. It looks like the second of them could be implemented internally and two others as library procedures.

The functions char-ready? and peek-char most probably are not working properly (and perhaps never were due to #3). Also consider bytewise versions peek-u8 and u8-ready?

Small but painful thing missing in R5RS is (read-line) procedure. This could be added as library function to save core code size.

UTF-8 is now working for strings and characters, but not for symbols, e.g.

(symbol->string 'bы)

Returns broken string with length of 1. This is because symbols are based on internal string representation, but this representation for unicode (with 4-byte chars for fast indexing) is not compatible with symbol-related internal functions.

This could be fixed if, instead, symbols have their own internal representation or rely on bytevector.

However it is not critical feature of course as the idea of using non-ascii characters in "symbols" (e.g. for parts of program code)
is strongly discouraged. So it more makes sense for consistency.