Hi -
I was working on a little improvement in UTF-8 conversion speed (so far it's about 150x faster for latin-1 text ;-) and for measuring the improvements was running a test that said:
strings := String allSubInstances. 1 to: strings size do:[:i| original := strings at: i. utf8 := original squeakToUtf8. copy := utf8 utf8ToSqueak. original = copy ifFalse:[self error: 'Encoding problem']. ].
When I ran this test it failed on each and every WideString instance. Digging into it, it seems that all of the WideStrings in Squeak have a language tag that is being supplied implicitly by the current LanguageEnvironment.
Questions: 1) From what it looks like right now there is no way to preserve that language tag through a UTF-8 conversion. Is this indeed the case or am I missing something? 2) Given that my language environment is being set to Latin-1, how should clients treat UTF-8 to provide the "proper" language tag? For example, I expected that a client be able to read and write UTF-8 text without implicitly providing that language tag. If that's the case, then how does one store these in common text files? (I could see how to do this for formatted text but not for "plain text files" without further attributation) 3) More generally asking, isn't the language tag here more of a "decorator" along the lines of text attributes? This would certainly model more closely the effect that I'm seeing here (some attributes are dropped by the squeak -> utf8 -> squeak conversion) *except* that I didn't expect any lossy conversion for strings (contrary to Text where dropping text attributes is obviously lossy).
Thanks for any help, - Andreas
On Jun 26, 2007, at 9:19 , Andreas Raab wrote:
Hi -
I was working on a little improvement in UTF-8 conversion speed (so far it's about 150x faster for latin-1 text ;-) and for measuring the improvements was running a test that said:
strings := String allSubInstances. 1 to: strings size do:[:i| original := strings at: i. utf8 := original squeakToUtf8. copy := utf8 utf8ToSqueak. original = copy ifFalse:[self error: 'Encoding problem']. ].
When I ran this test it failed on each and every WideString instance. Digging into it, it seems that all of the WideStrings in Squeak have a language tag that is being supplied implicitly by the current LanguageEnvironment.
Questions:
- From what it looks like right now there is no way to preserve
that language tag through a UTF-8 conversion. Is this indeed the case or am I missing something? 2) Given that my language environment is being set to Latin-1, how should clients treat UTF-8 to provide the "proper" language tag? For example, I expected that a client be able to read and write UTF-8 text without implicitly providing that language tag. If that's the case, then how does one store these in common text files? (I could see how to do this for formatted text but not for "plain text files" without further attributation) 3) More generally asking, isn't the language tag here more of a "decorator" along the lines of text attributes? This would certainly model more closely the effect that I'm seeing here (some attributes are dropped by the squeak -> utf8 -> squeak conversion) *except* that I didn't expect any lossy conversion for strings (contrary to Text where dropping text attributes is obviously lossy).
Nice catch. We had the discussion before, and this to me is another hint that we really should strip the language tag from Strings and move it to Text attributes. For rendering bare strings the default language could be taken from the current environment. The problem is, IIUC, that currently a lot of bare strings are passed around so it was simpler to just tag the language onto the string itself.
- Bert -
However, if you strip the language tag, you will run into very minor bugs with the A macron and a macron, because their encodings have been hijacked as CrossedX and EndOfRun in the CharacterScanner family (clever trick when Characters were 256). I searched how these damned characters could ever work in Squeak and Sophie, and found black magic was this language tag.
Andreas, maybe you could have a look at how RTF text are converted in SOphie, it seems to deal with language tag correctly, at least with extended latin characters.
Nicolas
Bert Freudenberg a écrit :
On Jun 26, 2007, at 9:19 , Andreas Raab wrote:
Hi -
I was working on a little improvement in UTF-8 conversion speed (so far it's about 150x faster for latin-1 text ;-) and for measuring the improvements was running a test that said:
strings := String allSubInstances. 1 to: strings size do:[:i| original := strings at: i. utf8 := original squeakToUtf8. copy := utf8 utf8ToSqueak. original = copy ifFalse:[self error: 'Encoding problem']. ].
When I ran this test it failed on each and every WideString instance. Digging into it, it seems that all of the WideStrings in Squeak have a language tag that is being supplied implicitly by the current LanguageEnvironment.
Questions:
- From what it looks like right now there is no way to preserve that
language tag through a UTF-8 conversion. Is this indeed the case or am I missing something? 2) Given that my language environment is being set to Latin-1, how should clients treat UTF-8 to provide the "proper" language tag? For example, I expected that a client be able to read and write UTF-8 text without implicitly providing that language tag. If that's the case, then how does one store these in common text files? (I could see how to do this for formatted text but not for "plain text files" without further attributation) 3) More generally asking, isn't the language tag here more of a "decorator" along the lines of text attributes? This would certainly model more closely the effect that I'm seeing here (some attributes are dropped by the squeak -> utf8 -> squeak conversion) *except* that I didn't expect any lossy conversion for strings (contrary to Text where dropping text attributes is obviously lossy).
Nice catch. We had the discussion before, and this to me is another hint that we really should strip the language tag from Strings and move it to Text attributes. For rendering bare strings the default language could be taken from the current environment. The problem is, IIUC, that currently a lot of bare strings are passed around so it was simpler to just tag the language onto the string itself.
- Bert -
nicolas cellier wrote:
However, if you strip the language tag, you will run into very minor bugs with the A macron and a macron, because their encodings have been hijacked as CrossedX and EndOfRun in the CharacterScanner family (clever trick when Characters were 256). I searched how these damned characters could ever work in Squeak and Sophie, and found black magic was this language tag.
Ah, how interesting. I wasn't even aware of that but it makes good sense. Which, in a sense, only emphasizes question 2) below given that the default Latin-1 environment doesn't seem to set a language code whatsoever.
Andreas, maybe you could have a look at how RTF text are converted in SOphie, it seems to deal with language tag correctly, at least with extended latin characters.
Good point. Unfortunately, I don't have the time to get into Sophie in detail (I was just trying to understand why UTF-8 conversion is lossy and what to do about it) but if someone would give me a primer on how Sophie deals with these issues I'd appreciate it.
Cheers, - Andreas
Nicolas
Bert Freudenberg a écrit :
On Jun 26, 2007, at 9:19 , Andreas Raab wrote:
Hi -
I was working on a little improvement in UTF-8 conversion speed (so far it's about 150x faster for latin-1 text ;-) and for measuring the improvements was running a test that said:
strings := String allSubInstances. 1 to: strings size do:[:i| original := strings at: i. utf8 := original squeakToUtf8. copy := utf8 utf8ToSqueak. original = copy ifFalse:[self error: 'Encoding problem']. ].
When I ran this test it failed on each and every WideString instance. Digging into it, it seems that all of the WideStrings in Squeak have a language tag that is being supplied implicitly by the current LanguageEnvironment.
Questions:
- From what it looks like right now there is no way to preserve that
language tag through a UTF-8 conversion. Is this indeed the case or am I missing something? 2) Given that my language environment is being set to Latin-1, how should clients treat UTF-8 to provide the "proper" language tag? For example, I expected that a client be able to read and write UTF-8 text without implicitly providing that language tag. If that's the case, then how does one store these in common text files? (I could see how to do this for formatted text but not for "plain text files" without further attributation) 3) More generally asking, isn't the language tag here more of a "decorator" along the lines of text attributes? This would certainly model more closely the effect that I'm seeing here (some attributes are dropped by the squeak -> utf8 -> squeak conversion) *except* that I didn't expect any lossy conversion for strings (contrary to Text where dropping text attributes is obviously lossy).
Nice catch. We had the discussion before, and this to me is another hint that we really should strip the language tag from Strings and move it to Text attributes. For rendering bare strings the default language could be taken from the current environment. The problem is, IIUC, that currently a lot of bare strings are passed around so it was simpler to just tag the language onto the string itself.
- Bert -
Andreas Raab wrote:
Good point. Unfortunately, I don't have the time to get into Sophie in detail (I was just trying to understand why UTF-8 conversion is lossy and what to do about it) but if someone would give me a primer on how Sophie deals with these issues I'd appreciate it.
The main point we did in Sophie was to build "true" unicode converters. The existing ones preserved some of the black magic, so when converting into unicode you actually didn't for some characters. So just looking at the new converters would be a good start. They can be found in the SOphie-RTF package.
Michael
MappingUnicodeTextConverter #() CP1250UnicodeTextConverter #() CP1251UnicodeTextConverter #() CP1252UnicodeTextConverter #() MacRomanUnicodeTextConverter #()
As Bert suggested, the Right Thing is to build a system on an assumption that bare String and Characters cannot be really displayed. For method source, the tag is encoded as the text property so they are retained. A XML-like (or whatever) format in UTF-8 for storing Squeak Text and use it almost always is the consecuence from it.
-- Yoshiki
At Tue, 26 Jun 2007 00:19:04 -0700, Andreas Raab wrote:
Hi -
I was working on a little improvement in UTF-8 conversion speed (so far it's about 150x faster for latin-1 text ;-) and for measuring the improvements was running a test that said:
strings := String allSubInstances. 1 to: strings size do:[:i| original := strings at: i. utf8 := original squeakToUtf8. copy := utf8 utf8ToSqueak. original = copy ifFalse:[self error: 'Encoding problem']. ].
When I ran this test it failed on each and every WideString instance. Digging into it, it seems that all of the WideStrings in Squeak have a language tag that is being supplied implicitly by the current LanguageEnvironment.
Questions:
- From what it looks like right now there is no way to preserve that
language tag through a UTF-8 conversion. Is this indeed the case or am I missing something? 2) Given that my language environment is being set to Latin-1, how should clients treat UTF-8 to provide the "proper" language tag? For example, I expected that a client be able to read and write UTF-8 text without implicitly providing that language tag. If that's the case, then how does one store these in common text files? (I could see how to do this for formatted text but not for "plain text files" without further attributation) 3) More generally asking, isn't the language tag here more of a "decorator" along the lines of text attributes? This would certainly model more closely the effect that I'm seeing here (some attributes are dropped by the squeak -> utf8 -> squeak conversion) *except* that I didn't expect any lossy conversion for strings (contrary to Text where dropping text attributes is obviously lossy).
Thanks for any help,
- Andreas
Yoshiki Ohshima wrote:
As Bert suggested, the Right Thing is to build a system on an assumption that bare String and Characters cannot be really displayed.
I'm sure it is, but unfortunately we don't have the time to do the Right Thing since we need to get a product out the door ;-)
For method source, the tag is encoded as the text property so they are retained. A XML-like (or whatever) format in UTF-8 for storing Squeak Text and use it almost always is the consecuence from it.
Thanks. So if I hear you correctly you are recommending to preserve the language tag via additional attributes. Is that correct?
Cheers, - Andreas
-- Yoshiki
At Tue, 26 Jun 2007 00:19:04 -0700, Andreas Raab wrote:
Hi -
I was working on a little improvement in UTF-8 conversion speed (so far it's about 150x faster for latin-1 text ;-) and for measuring the improvements was running a test that said:
strings := String allSubInstances. 1 to: strings size do:[:i| original := strings at: i. utf8 := original squeakToUtf8. copy := utf8 utf8ToSqueak. original = copy ifFalse:[self error: 'Encoding problem']. ].
When I ran this test it failed on each and every WideString instance. Digging into it, it seems that all of the WideStrings in Squeak have a language tag that is being supplied implicitly by the current LanguageEnvironment.
Questions:
- From what it looks like right now there is no way to preserve that
language tag through a UTF-8 conversion. Is this indeed the case or am I missing something? 2) Given that my language environment is being set to Latin-1, how should clients treat UTF-8 to provide the "proper" language tag? For example, I expected that a client be able to read and write UTF-8 text without implicitly providing that language tag. If that's the case, then how does one store these in common text files? (I could see how to do this for formatted text but not for "plain text files" without further attributation) 3) More generally asking, isn't the language tag here more of a "decorator" along the lines of text attributes? This would certainly model more closely the effect that I'm seeing here (some attributes are dropped by the squeak -> utf8 -> squeak conversion) *except* that I didn't expect any lossy conversion for strings (contrary to Text where dropping text attributes is obviously lossy).
Thanks for any help,
- Andreas
Andreas,
For method source, the tag is encoded as the text property so they are retained. A XML-like (or whatever) format in UTF-8 for storing Squeak Text and use it almost always is the consecuence from it.
Thanks. So if I hear you correctly you are recommending to preserve the language tag via additional attributes. Is that correct?
The short answer is, yes.
-- Yoshiki
What's the status of these patches? Seaside shows a measurable speed drop when doing utf-8 encoding/decoding so we'd be more than willing to test them. We don't care about the stripping of language tags, we are fine with the unification aspect of unicode.
Cheers Philippe
2007/6/26, Andreas Raab andreas.raab@gmx.de:
Hi -
I was working on a little improvement in UTF-8 conversion speed (so far it's about 150x faster for latin-1 text ;-) and for measuring the improvements was running a test that said:
strings := String allSubInstances. 1 to: strings size do:[:i| original := strings at: i. utf8 := original squeakToUtf8. copy := utf8 utf8ToSqueak. original = copy ifFalse:[self error: 'Encoding problem']. ].
When I ran this test it failed on each and every WideString instance. Digging into it, it seems that all of the WideStrings in Squeak have a language tag that is being supplied implicitly by the current LanguageEnvironment.
Questions:
- From what it looks like right now there is no way to preserve that
language tag through a UTF-8 conversion. Is this indeed the case or am I missing something? 2) Given that my language environment is being set to Latin-1, how should clients treat UTF-8 to provide the "proper" language tag? For example, I expected that a client be able to read and write UTF-8 text without implicitly providing that language tag. If that's the case, then how does one store these in common text files? (I could see how to do this for formatted text but not for "plain text files" without further attributation) 3) More generally asking, isn't the language tag here more of a "decorator" along the lines of text attributes? This would certainly model more closely the effect that I'm seeing here (some attributes are dropped by the squeak -> utf8 -> squeak conversion) *except* that I didn't expect any lossy conversion for strings (contrary to Text where dropping text attributes is obviously lossy).
Thanks for any help,
- Andreas
squeak-dev@lists.squeakfoundation.org