Re: RFC Wide Characters in I/O

[Tobias Burnus <burnus at net-b dot de>]

Jerry DeLisle wrote:
> If we are given a wide character to transfer as unformatted, we simply 
> transfer all the bytes as is. (I will confirm with the standard on this.)
I think this OK as the standard does not define the on-disk format; 
additionally, ENCODING= can only be given for formatted I/O. One should 
make sure that the endianess is properly honored. As a wide character is 
simply an INTEGER(4) one can simply write them as integer(4) [array] - 
what they actually are.

> If the user has specified an ENCODING="default" and the kind is 1, we 
> do what we do now and transfer as 8bit (mostly ASCII).
OK. This automatically allows to support 8-bit encodings and is backward 
compatible.

[still formatted and encoding='default':]
> If kind is greater than 1, I suggest we transfer each byte as is. So 
> for kind=4, we would transfer 4 bytes. This would enable doing some 
> packed 4x1 byte character stuff.  I think the standard would allow 
> this and thats why "default" is so loosely defined.
I think for I/O, almost everything is allowed by the standard. Note that 
the standard distinguishes between DEFAULT and ASCII characters. In case 
of gfortran, they are the same and thus the better-defined ASCII applies.

Somehow I think your suggestion is not necessarily the best one. For

write(foo,'(a)') ucs4_"Hello"

I expect that it writes 5 bytes and not 20 bytes to the unit "foo" or 
internal file "foo". Regarding internal files one finds in the standard:

"An input/output list shall not contain an item of nondefault character 
type if the input/output statement specifies an internal file of default 
character type. An input/output list shall not contain an item of 
nondefault character type other than ISO 10646 or ASCII character type 
if the input/output statement specifies an internal file of ISO 10646 
character type. An input/output list shall not contain a character item 
of any character type other than ASCII  character type if the 
input/output statement specifies an internal file of ASCII character type."

If I read this correctly, the standard requires likewise. Thus I think 
it is OK to give an error if the character is not representable in a 
byte. (Thus char < 128 or char < 256 should be rejected.) Alternatively, 
one can convert these characters into a "?".

For files one is more flexible, but still I think being able to write a 
normal ASCII string without needing to assign the ucs4-kind string to a 
default-kind variable first is useful. Again either an error or 
conversion to "?" could be a solution.


For the sake of completeness, the following can also be found in the 
standard:

"During input from a Unicode file,
 (1) characters in the record that correspond to an ASCII character 
variable shall have a position in the ISO 10646 character type collating 
sequence of 127 or less, and
 (2) characters in the record that correspond to a default character 
variable shall be representable in the default character type.

During input from a non-Unicode file,
 (1) characters in the record that correspond to a character variable 
shall have the kind of the character variable, and
 (2) characters in the record that correspond to a numeric or logical 
variable shall be of default character type.

During output to a Unicode file, all characters transmitted to the 
record are of ISO 10646 character type. If a character input/output list 
item or character string edit descriptor contains a character that is 
not representable in the ISO 10646 character type, the result is 
processor-dependent.
 During output to a non-Unicode file, characters transmitted to the 
record as a result of processing a character string edit descriptor or 
as a result of evaluating a numeric, logical, or default character data 
entity, are of type default character."

I think writing a wide-char variable into a default-kind internal file 
should give an error for characters which cannot fit into the variable. 
(One can argue whether only ASCII characters or also ISO-8859-1 (Latin1) 
characters fit.)


> (Note: It is my understanding, in reading about UTF-8, that it is 
> internally represented by a fixed width 4 byte hexidecimal number 
> which when transfered to an external file, is translated on the fly 
> into a variable width encoding.  When reading back in, it is 
> translated back to the 4 byte code.)
I'm not 100% sure whether I read this correctly, but the UTF-8 
characters should be represented as UCS-4 (integer(4)) in the library 
until the record is finally written.

> If the user has specified an ENCODING="UTF-8" and the kind is less 
> than 4, strictly speaking, thats an error, but we can give a warning 
> and assume a truncation.
I guess, you are talking about READING from a file; and (see quote 
above) this is valid as long as the characters are representable. Thus I 
would go again for "?" or for an error.

Tobias

PS: I think it is a matter of taste, but I slightly favour printing a 
"?" to giving a run-time error.