Probably fixnum has the same problem. Even Non-Typed-Racket has a under the hood optimization for loops with fixnums, but it doesn't detect the corner case (in-range 1073741824).
Just an idea: Is it possible to change (in-range 1073741824) automatically to (in-range-inclusive 1073741823) and allowd additional optimizations? Nevertheless, (in-range 256) is probably more usual. Gustavo On Wed, Feb 18, 2015 at 4:59 PM, Neil Toronto <neil.toro...@gmail.com> wrote: > TR isn't really good about finding precise types for range sequence > elements. The macro-expanded code is too complicated. > > When I write a loop with an index that needs to have a specific, narrow > type, I either use an assertion, as in > > (for/list : (Listof Byte) ([b (in-range 256)]) > (assert b byte?)) > > or loop using the next wider type and always compare the loop index with a > value that has the narrow type: > > (reverse > (let loop : (Listof Byte) ([b : Index 0] > [bs : (Listof Byte) empty]) > (cond [(<= b 255) (loop (+ b 1) (cons b bs))] > [else bs]))) > > The second method is about 10% faster. > > Neil ⊥ > > > On 02/18/2015 02:38 PM, John Clements wrote: >> >> I want to write this: >> >> (for/list ([b (in-range 256)]) >> b) >> >> ... and get back a list of bytes. >> >> That is, I want TR to see that the number 256 is one greater than the >> largest byte, and therefore that b must be of type byte. I see that >> this dips way too close to undecidability in the general case, so I >> would instead perhaps expect to see something like an 'in-bytes-range' >> form that is known to produce only bytes. >> >> I looked for such a thing, but didn't find it. Naturally, I could >> simply jam a run-time check in at the top of the body, something like >> 'ensure-bytes', but it seems unnecessary. >> >> Am I missing something obvious? >> >> Thanks! >> >> John >> >> >> >> ____________________ >> Racket Users list: >> http://lists.racket-lang.org/users >> > > ____________________ > Racket Users list: > http://lists.racket-lang.org/users ____________________ Racket Users list: http://lists.racket-lang.org/users
