On Tue, Aug 24, 2021 at 09:03:30AM -0400, Viktor Dukhovni wrote: I originally wrote:

...

...

Is there any GHC syntax for constructing a primitive string literal with a known (not hand coded) byte count? With `"some bytes"#` I get just the `Addr#` pointer, but not the size.

If there's nothing available, would it be reasonable to introduce a new syntax? Perhaps:

"some bytes"## :: (# Addr#, Int# #)

But neglected to mention that I knew about `cstringLength#`, but found it wanting, because it does not support octet-strings with embedded NUL characters:

Sadly, that does not work when the primitive octet string contains internal NUL bytes.

λ> :set -package ghc-prim λ> :set -XMagicHash λ> import GHC.CString λ> import GHC.Int λ> λ> I# (cstringLength# "foobar\xa0"#) 7 λ> I# (cstringLength# "foo\0bar\xa0"#) 3

If there isn't some other extant work-around, any feedback on my proposal of a new syntax for a primitive unboxed (address, length) pair: "some bytes"## :: (# Addr#, Int# #) -- Viktor.

On Wed, Aug 25, 2021 at 07:05:58PM +0300, Oleg Grenrus wrote:

The newew proposal [1] is tagged as "needs revision". It doesn't include(# Int#, Addr# #), but those are easy to get from ByteArray# which has negligible overhead. [...] [1] https://github.com/ghc-proposals/ghc-proposals/pull/292

Yes, ByteArray# literals would work just as well for my needs. The one thing that's missing, from the proposed variants: Rather than adding new syntax, this proposal leverages an existing GHC extension: QuasiQuotes. Rather than using TemplateHaskell, these quasiquoters would be built in to the compiler. Here are some examples of ByteArray# literals under this scheme: [octets|fe01bce8|] -- ByteArray# (four bytes) [utf8|Araña|] -- ByteArray# (UTF-8) [utf16|Araña|] -- ByteArray# (UTF-16, native endian) [utf16le|Araña|] -- ByteArray# (UTF-16, little endian) [utf16be|Araña|] -- ByteArray# (UTF-16, big endian) is a syntax for octet-strings that does not force hex encoding of every byte, thus something along the lines of: [octetstr|foo%A0bar|] -- ByteArray# (seven bytes) The "%hh" hex octet could be "\hh" or "\xhh", ... whatever is deemed sufficiently natural/readable (perhaps "foo\xA0\&bar" for consistency with Haskell strings?). The "\xhh" form would be familiar to Python users: >>> x = b'foo\xA0bar' >>> len(x) 7 >>> x[3] 160 So, I support the proposal, even though quasi-quoters are more bulky than "somebytes"##, they have the advantage of supporting multiple variant formats. I might be tempted to use "octets" for the non-hex form with "%" or other escapes, and "hexstr" (or similar) for the hex form. -- Viktor.