Well, the alternative to exploiting polymorphy like in the patch would
be to have a nearly identical bunch of "TextConverter" classes that
can read from binary streams and answer characters (like the Zinc UTF8
"Encoder" that also happens to have decoding methods). But most of the
code would be duplicated with the existing TextConverters, so I think
it is the worse solution.

IMHO, a fundamental thing such as a text encoding interpreter that is
shipped with a standard library should not be restricted by "it was
only designed to be used with consumer xyz". The TextConverter
hierarchy is neither marked as internal/private in Multilingual by a
category, nor by any comments, nor do the class names suggest coupling
to the MultiByte*Streams.

If I were to commit this thing from scratch, I would even drop the
support for returning the bytes as they are if the source stream
#isBinary, because it has nothing to do with text conversion. Instead,
do not use a TextConverter, if you do not want the bytes converted.
But, as Levente already wrote, we can not do this without a larger
refactoring and compatibility issues.

Best regards,
Jakob

2017-01-23 19:24 GMT+01:00 H. Hirzel <hannes.hirzel at gmail.com>:
> Interesting in this context  the UTF8 decoding implementation of Pharo
> 5 ZnUTF8Encoder (an alternative to UTF8TextConverter it seems)
>
> ZnUTF8Encoder>>
> nextFromStream: stream
>         | code byte next |
>         (byte := stream next) < 128
>                 ifTrue: [ ^ Character codePoint: byte ].
>         (byte bitAnd: 2r11100000) == 2r11000000
>                 ifTrue: [
>                         code := byte bitAnd: 2r00011111.
>                         ((next := stream next ifNil: [ self errorIncomplete ]) bitAnd:
> 2r11000000) == 2r10000000
>                                 ifTrue: [ code := (code bitShift: 6) + (next bitAnd: 2r00111111) ]
>                                 ifFalse: [ ^ self errorIllegalContinuationByte ].
>                         code < 128 ifTrue: [ self errorOverlong ].
>                         ^ Character codePoint: code ].
>         (byte bitAnd: 2r11110000) == 2r11100000
>                 ifTrue: [
>                         code := byte bitAnd: 2r00001111.
>                         2 timesRepeat: [
>                                 ((next := stream next ifNil: [ self errorIncomplete ]) bitAnd:
> 2r11000000) == 2r10000000
>                                         ifTrue: [ code := (code bitShift: 6) + (next bitAnd: 2r00111111) ]
>                                         ifFalse: [ ^ self errorIllegalContinuationByte ] ].
>                         code < 2048 ifTrue: [ self errorOverlong ].
>                         code = 65279 "Unicode Byte Order Mark" ifTrue: [
>                                 stream atEnd ifTrue: [ self errorIncomplete ].
>                                 ^ self nextFromStream: stream ].
>                         ^ Character codePoint: code ].
>         (byte bitAnd: 2r11111000) == 2r11110000
>                 ifTrue: [
>                         code := byte bitAnd: 2r00000111.
>                         3 timesRepeat: [
>                                 ((next := stream next ifNil: [ self errorIncomplete ]) bitAnd:
> 2r11000000) == 2r10000000
>                                         ifTrue: [ code := (code bitShift: 6) + (next bitAnd: 2r00111111) ]
>                                         ifFalse: [ ^ self errorIllegalContinuationByte ] ].
>                         code < 65535 ifTrue: [ self errorOverlong ].
>                         ^ Character codePoint: code ].
>         self errorIllegalLeadingByte
>
> On 1/23/17, Levente Uzonyi <leves at caesar.elte.hu> wrote:
>> The Pharo version seems to be the Squeak version optimized for
>> VisualWorks (ifNil: -> isNil ifTrue:).
>>
>> Levente
>>
>> On Mon, 23 Jan 2017, H. Hirzel wrote:
>>
>>> Below as a comparison the version in Pharo 5.0.
>>>
>>> Noteworthy to say is that one can not speak about characters in an
>>> UTF8 encoded stream which is read byte by byte until one has examined
>>> the bytes.
>>>
>>> So if I read the first thing it is actually a byte. Then I can examine
>>> if it is a one-byte character and then return the character. Then I go
>>> for the next byte. If it indicates that we have a two byte encoded
>>> UTF8 character then I can return the character.
>>>
>>> So I should have
>>>
>>> byte1 := aStream basicNext.
>>>
>>> ... check if we have a one byte character, if yes return the character
>>>
>>> byte2 := aStream basicNext.
>>>
>>> ... check if we have a two byte character, if yes return the character
>>>
>>> byte3 := aStream basicNext.
>>>
>>> ... check if we have a three byte character, if yes return the character
>>>
>>>
>>> byte4 := aStream basicNext.
>>>
>>> ... check if we have a four byte character, if yes return the character
>>>
>>>
>>>
>>>
>>>
>>> nextFromStream: aStream
>>>      | character1 value1 character2 value2 unicode character3 value3
>>> character4 value4 |
>>>      aStream isBinary
>>>              ifTrue: [ ^ aStream basicNext ].
>>>      character1 := aStream basicNext.
>>>      character1 isNil
>>>              ifTrue: [ ^ nil ].
>>>      value1 := character1 asciiValue.
>>>      value1 <= 127
>>>              ifTrue: [
>>>                      "1-byte character"
>>>                      ^ character1 ]. "at least 2-byte character"
>>>      character2 := aStream basicNext.
>>>      character2 isNil
>>>              ifTrue: [ ^ self errorMalformedInput ].
>>>      value2 := character2 asciiValue.
>>>      (value1 bitAnd: 16rE0) = 192
>>>              ifTrue: [ ^ Unicode value: ((value1 bitAnd: 31) bitShift: 6) +
>>> (value2 bitAnd: 63) ].       "at least 3-byte character"
>>>      character3 := aStream basicNext.
>>>      character3 isNil
>>>              ifTrue: [ ^ self errorMalformedInput ].
>>>      value3 := character3 asciiValue.
>>>      (value1 bitAnd: 16rF0) = 224
>>>              ifTrue: [ unicode := ((value1 bitAnd: 15) bitShift: 12) + ((value2
>>> bitAnd: 63) bitShift: 6) + (value3 bitAnd: 63) ].
>>>      (value1 bitAnd: 16rF8) = 240
>>>              ifTrue: [
>>>                      "4-byte character"
>>>                      character4 := aStream basicNext.
>>>                      character4 isNil
>>>                              ifTrue: [ ^ self errorMalformedInput ].
>>>                      value4 := character4 asciiValue.
>>>                      unicode := ((value1 bitAnd: 16r7) bitShift: 18) + ((value2 bitAnd:
>>> 63) bitShift: 12)
>>>                              + ((value3 bitAnd: 63) bitShift: 6) + (value4 bitAnd: 63) ].
>>>      unicode isNil
>>>              ifTrue: [ ^ self errorMalformedInput ].
>>>      unicode > 16r10FFFD
>>>              ifTrue: [ ^ self errorMalformedInput ].
>>>      unicode = 16rFEFF
>>>              ifTrue: [ ^ self nextFromStream: aStream ].
>>>      ^ Unicode value: unicode
>>>
>>>
>>>
>>>
>>>
>>> On 1/22/17, Tobias Pape <Das.Linux at gmx.de> wrote:
>>>>
>>>> On 22.01.2017, at 16:10, Levente Uzonyi <leves at caesar.elte.hu> wrote:
>>>>
>>>>> On Fri, 20 Jan 2017, Tobias Pape wrote:
>>>>>
>>>>>>
>>>>>> On 19.01.2017, at 23:30, Levente Uzonyi <leves at caesar.elte.hu> wrote:
>>>>>>
>>>>>>> On Thu, 19 Jan 2017, Tobias Pape wrote:
>>>>>>>> Thanks Jacob.
>>>>>>>> Any objections here I put this into trunk?
>>>>>>> Yep. TextConverters are intended to work with MultiByte*Streams only.
>>>>>>
>>>>>> Didn't know that.
>>>>>>
>>>>>>> Therefore #basicNext is expected to return a Character, provided the
>>>>>>> stream is not binary. This is why the #isBinary check is the first
>>>>>>> thing
>>>>>>> the method does.
>>>>>>
>>>>>> I see. however, using asInteger sounds more reasonable _even though_ it
>>>>>> is a character. Said bluntly, the responsibility of the TextConverter
>>>>>> is
>>>>>> to make Characters from that bloody numbers in that stream.
>>>>>> I was confused to see that asciiValue returns something >127 in the
>>>>>> first
>>>>>> place.
>>>>>
>>>>> #asInteger does the same thing as #asciiValue. While #asciiValue doesn't
>>>>> do what you would expect it to do, it has the advantage to clearly mark
>>>>> the class of the receiver (in this case).
>>>>>
>>>>
>>>> Yes, and that's exactly why we should use #asInteger. To _not_ limit the
>>>> receiver.
>>>> Because the receiver isn't actually a Character, but some number, encoded
>>>> in
>>>> a Character, whose meaning is to be determined by
>>>> this very method.
>>>>
>>>> Also, how do we know that _basic_Next will always return a Character?
>>>> (Yes, I know there's a binary check, but doesn't that only say something
>>>> about #next, not #basicNext?)
>>>>
>>>>
>>>>>>
>>>>>>> If there are plans to make TextConverters work with more general
>>>>>>> streams, then I persume these changes won't be enough.
>>>>>>
>>>>>> Clearly.
>>>>>> But isn't this a step in the right direction?
>>>>>
>>>>> Yes and no. There are at least two ways to go:
>>>>>
>>>>> 1. Enhance the current stream library, even at the cost of breaking
>>>>> things.
>>>>> A patch here and there won't work. There are fundamental changes
>>>>> required,
>>>>> like stackable streams, to make it desirable to use it over other
>>>>> libraries.
>>>>>
>>>>> 2. Integrate an existing stream library with better features (e.g.
>>>>> Xtreams)
>>>>> If we were to do this, we could gradually migrate existing code to the
>>>>> new
>>>>> library, and finally make the current stream library unloadable.
>>>>
>>>> I like the idea of Xtreams, but I also like going baby steps.
>>>>
>>>> The changes here help at least one person, won't hurt others and seem
>>>> future
>>>> proof.
>>>> So?
>>>>
>>>> Best regards
>>>>     -Tobias
>>>>
>>>>
>>>>
>>>>> Levente
>>>>>
>>>>>>
>>>>>>> Levente
>>>>>>>> Looks good from here.
>>>>>>>> Best regards
>>>>>>>>         -Tobias
>>>>>>>> On 19.01.2017, at 17:14, commits at source.squeak.org wrote:
>>>>>>>>> A new version of Multilingual was added to project The Inbox:
>>>>>>>>> http://source.squeak.org/inbox/Multilingual-jr.218.mcz
>>>>>>>>> ==================== Summary ====================
>>>>>>>>> Name: Multilingual-jr.218
>>>>>>>>> Author: jr
>>>>>>>>> Time: 19 January 2017, 5:14:23.763655 pm
>>>>>>>>> UUID: 36416c42-a4b4-554f-8203-aba25eee794f
>>>>>>>>> Ancestors: Multilingual-tfel.217
>>>>>>>>> support 'iso-8859-1' and do not let UTF8TextConverter expect that
>>>>>>>>> its
>>>>>>>>> input stream  returns Characters from basicNext
>>>>>>>>> A stream implementation might always return bytes from basicNext and
>>>>>>>>> expect the conversion to Character to be done solely by the
>>>>>>>>> TextConverter, so use asInteger instead of asciiValue to support
>>>>>>>>> both
>>>>>>>>> cases. Convert back with asCharacter.
>>>>>>>>> =============== Diff against Multilingual-tfel.217 ===============
>>>>>>>>> Item was changed:
>>>>>>>>> ----- Method: Latin1TextConverter class>>encodingNames (in category
>>>>>>>>> 'utilities') -----
>>>>>>>>> encodingNames +        ^ #('latin-1' 'latin1' 'iso-8859-1') copy.
>>>>>>>>> -      ^ #('latin-1' 'latin1') copy.
>>>>>>>>> !
>>>>>>>>> Item was changed:
>>>>>>>>> ----- Method: UTF8TextConverter>>nextFromStream: (in category
>>>>>>>>> 'conversion') -----
>>>>>>>>> nextFromStream: aStream
>>>>>>>>>
>>>>>>>>>        | char1 value1 char2 value2 unicode char3 value3 char4 value4 |
>>>>>>>>>        aStream isBinary ifTrue: [^ aStream basicNext].
>>>>>>>>>        char1 := aStream basicNext.
>>>>>>>>>        char1 ifNil:[^ nil].
>>>>>>>>> +      value1 := char1 asInteger.
>>>>>>>>> -      value1 := char1 asciiValue.
>>>>>>>>>        value1 <= 127 ifTrue: [
>>>>>>>>>                "1-byte char"
>>>>>>>>> +              ^ char1 asCharacter
>>>>>>>>> -              ^ char1
>>>>>>>>>        ].
>>>>>>>>>
>>>>>>>>>        "at least 2-byte char"
>>>>>>>>>        char2 := aStream basicNext.
>>>>>>>>> +      char2 ifNil:[^self errorMalformedInput: (String with: char1
>>>>>>>>> asCharacter)].
>>>>>>>>> +      value2 := char2 asInteger.
>>>>>>>>> -      char2 ifNil:[^self errorMalformedInput: (String with: char1)].
>>>>>>>>> -      value2 := char2 asciiValue.
>>>>>>>>>
>>>>>>>>>        (value1 bitAnd: 16rE0) = 192 ifTrue: [
>>>>>>>>>                ^ Unicode value: ((value1 bitAnd: 31) bitShift: 6) + (value2
>>>>>>>>> bitAnd:
>>>>>>>>> 63).
>>>>>>>>>        ].
>>>>>>>>>
>>>>>>>>>        "at least 3-byte char"
>>>>>>>>>        char3 := aStream basicNext.
>>>>>>>>> +      char3 ifNil:[^self errorMalformedInput: (String with: char1
>>>>>>>>> asCharacter with: char2 asCharacter)].
>>>>>>>>> +      value3 := char3 asInteger.
>>>>>>>>> -      char3 ifNil:[^self errorMalformedInput: (String with: char1 with:
>>>>>>>>> char2)].
>>>>>>>>> -      value3 := char3 asciiValue.
>>>>>>>>>        (value1 bitAnd: 16rF0) = 224 ifTrue: [
>>>>>>>>>                unicode := ((value1 bitAnd: 15) bitShift: 12) + ((value2 bitAnd:
>>>>>>>>> 63)
>>>>>>>>> bitShift: 6)
>>>>>>>>>                                + (value3 bitAnd: 63).
>>>>>>>>>        ].
>>>>>>>>>
>>>>>>>>>        (value1 bitAnd: 16rF8) = 240 ifTrue: [
>>>>>>>>>                "4-byte char"
>>>>>>>>>                char4 := aStream basicNext.
>>>>>>>>> +              char4 ifNil:[^self errorMalformedInput: (String with: char1
>>>>>>>>> asCharacter with: char2 asCharacter with: char3 asCharacter)].
>>>>>>>>> +              value4 := char4 asInteger.
>>>>>>>>> -              char4 ifNil:[^self errorMalformedInput: (String with: char1
>>>>>>>>> with:
>>>>>>>>> char2 with: char3)].
>>>>>>>>> -              value4 := char4 asciiValue.
>>>>>>>>>                unicode := ((value1 bitAnd: 16r7) bitShift: 18) +
>>>>>>>>>                                        ((value2 bitAnd: 63) bitShift: 12) +
>>>>>>>>>                                        ((value3 bitAnd: 63) bitShift: 6) +
>>>>>>>>>                                        (value4 bitAnd: 63).
>>>>>>>>>        ].
>>>>>>>>> +      unicode ifNil:[^self errorMalformedInput: (String with: char1
>>>>>>>>> asCharacter with: char2 asCharacter with: char3 asCharacter)].
>>>>>>>>> -      unicode ifNil:[^self errorMalformedInput: (String with: char1
>>>>>>>>> with:
>>>>>>>>> char2 with: char3)].
>>>>>>>>>        unicode > 16r10FFFD ifTrue: [
>>>>>>>>> +              ^self errorMalformedInput: (String with: char1 asCharacter with:
>>>>>>>>> char2 asCharacter with: char3 asCharacter).
>>>>>>>>> -              ^self errorMalformedInput: (String with: char1 with: char2 with:
>>>>>>>>> char3).
>>>>>>>>>        ].
>>>>>>>>>
>>>>>>>>>        unicode = 16rFEFF ifTrue: [^ self nextFromStream: aStream].
>>>>>>>>>        ^ Unicode value: unicode.
>>>>>>>>> !
>>>>
>>>>
>>>>
>>
>>
>