Best way to write some bytes to a Stream

[aarroyoc]

If I have some bytes in a normal list, what is the best way to write them to a stream? I suppose there must be something better than just calling put_byte/2 N times.

hex_bytes("ff", Bs), open("out", write, S), phrase_to_stream(seq(Bs), S).

This example fails, because phrase_to_stream/2 doesn't seem to like byte lists.
Using the internal $put_chars/2 which says it works for binary streams, doesn't seem to work, but it silently fails.

Is there something am I missing or do we need to add a new predicate?

[triska]

The best way is to represent bytes as a list of characters with codes in range 0..255. In this way, you can use the efficient internal representation of Scryer Prolog. You can use for example format(Stream, "~s", [Cs]) to write these characters to a binary stream. format/3 will automatically interpret them as bytes if Stream is binary.

You should not use '$put_chars/'2 directly. You can convert lists of bytes (integers in 0..255) Bs to lists of characters Cs with char_code/2: maplist(char_code, Cs, Bs). Ideally, this is avoided, and characters are used throughout.

See also #1072.

[aarroyoc]

Thanks! Yes, in the end I stopped using the byte list. However, this char_code/2 trick was interesting too

[triska]

Using phrase_to_stream/2 is also possible, and preferable over format/3 especially if you already have a DCG to describe the output (as a list of characters).

For efficiency, the most important point is to write as many bytes as possible "at once", and hence to avoid many successive writes of individual bytes. If you use phrase_to_stream/2, then a future implementation improvement to Scryer Prolog will even avoid the manifestation of the entire list in memory, by writing chunks of characters as they become known, and reclaiming them immediately.

[UWN]

What is the type of these byte-chars? They are a subset of character. I am a bit hesitant as this implies that all code points from '\0\' up to ÿ have to be valid characters. This is the case for Scryer, but not necessarily the case for every Prolog. In particular the zero character is not present everywhere.

[triska]

Hm, is this not a good argument (i.e., supporting binary data compactly) to add support for '\0\' in systems that do not yet have it? Is there any reason or advantage to insist on not supporting it in such systems?

[UWN]

In any case, you would need some type/domain. And, there are systems that just support only characters up to ASCII ~.

[triska]

What about byte_char, as mentioned in #1309?

[UWN]

@triska: what about octet_char?

[triska]

Since we already have character and chars, maybe the analogy is octet_character and octet_chars?

[UWN]

In any case, _character is definitely better than _char

[triska]

I have filed #1587 to add this!