Dan Kogai <dankogai(_at_)dan(_dot_)co(_dot_)jp> writes:
B. When you have a partial encoding consume as much as you can
and leave "string" with what is partial.
e.g.
abcd<ESC-to-jis0208>cdefghijklmn<ESC-to -ascii>opqrstu....
^- buffer boundary
Then you return translation of
"abcd<ESC-to-jis0208>cdefghijklmn"
and set "remains" to "<Esc-to"
so that :encoding can append "-ascii>opqrstu....
One of many reasons that programmers dislike 7bit ISO-2022 is exactly
how to handle case B -- how to split the buffer in the middle When
handling 7bit ISO-2022, YOU ARE NOT SUPPORSED TO SPLIT THE BUFFER BY
LENGTH.
But perlio.c and so :encoding does buffer by size right now.
So question is how to proceed from where we are.
Of course that causes the problem for large files and even
worse, network streams.
Which is what caused the problem to come to light.
But fortunately, 7bit ISO-2022 has one safety
net for that solution; IT ALWAYS REVERTS TO ASCII BEFORE CONTROL
CHARACTERS, including CRLF. So if you need it you can safely split
buffer line by line. A script
binmode(STDOUT, ":utf8");
while(<>){
print Encode:decode("iso-2022-jp", $_);
}
is completely safe because $_ is guaranteed to start in ASCII and end in
ASCII.
So what needs to happen is that :encoding(iso-2022-jp) needs to push
a layer underneath itself which makes sure a whole "line" is available.
You still need to clear the string to let me know that all is well.
Or we can mess with the decoder to "fake" it as you suggested.
(Tcl manages it ...)
Check RFC 1468 (http://www.ietf.org/rfc/rfc1468.txt and others). It is
not as complicated as it sounds.
I read that back before someone else implmented our first escape encoding.
If you cannot do that then don't return or consume anything
so :encoding can keep appending till you have whole file but that
is going to be very memory hungry.
As I said, "if you are worried about memory, just use line buffer" is
the answer.
So we need some way of telling from an encoding object (e.g.
an attribute or a method call) that it needs line buffering
so that :encoding layer can take the appropriate steps.
Other encodings are subject to this boundary problem -- and solution.
Arabic and Hebrew (BIDI boundary), Thai (word boundary), Hangul (for
decomposed form), you name it. But very fortunately for all these,
legacy encodings for those are all designed so that you can rely on CRLF
to split the stream.
Dan the Encode Maintainer.
--
Nick Ing-Simmons
http://www.ni-s.u-net.com/
