A few weeks ago, I mentioned 3 big things I'd like to see for merging Zope 2 and Zope 3: - Common Publisher - Common Security frameworks - Common ZPT implementations I forgot a very important need: - Common approach to Unicode In particular, In Zope 3, all text is stored and managed as Unicode. The publisher decodes request data and encodes response data. The vast majority of application and library code can ignore encoding issues. (The exceptions are applications and frameworks that need to exhange text with non-Unicode-aware external systems.) This has provided great simplifications and allowed us to avoid common pitfals from mixing Unicode and encoded text. We need to migrate Zope 2 to use a similar strategy. We need volunteers to brainstorm how this can be done and make one or more proposals. This is likely a prerequisite for finishing the publisher and ZPT work. Jim -- Jim Fulton mailto:jim@zope.com Python Powered! CTO (540) 361-1714 http://www.python.org Zope Corporation http://www.zope.com http://www.zope.org
Jim Fulton wrote:
I forgot a very important need:
- Common approach to Unicode
In particular, In Zope 3, all text is stored and managed as Unicode. The publisher decodes request data and encodes response data. The vast majority of application and library code can ignore encoding issues. (The exceptions are applications and frameworks that need to exhange text with non-Unicode-aware external systems.) This has provided great simplifications and allowed us to avoid common pitfals from mixing Unicode and encoded text.
We need to migrate Zope 2 to use a similar strategy. We need volunteers to brainstorm how this can be done and make one or more proposals. This is likely a prerequisite for finishing the publisher and ZPT work.
This is definitely a scary topic, and I speak from years of experience with Zope 2 unicode here. This sounds like a very hard transition that would touch *a lot* of code in non-Zope 2 core. How do you envision all the form inputs to suddenly produce unicode strings, for instance? We've struggled hard with Formulator to make it work with unicode for instance (and still it's buggy, as I wanted to support the non-unicode scenarios too). I can imagine any system in Zope that uses forms at all would need to be touched. I'll volunteer to help brainstorm on this, but right now my brainstorm is only very dark and full of lightning. Anyway, in some basics, Zope 2 does have an approach to unicode for *output* that's fairly similar to Zope 3's: if you use unicode strings your entire output (including page templates) will be unicode (if you don't mix with non-unicode non-ascii strings..). Then the response encoding setting is read and everything is transformed once to unicode text. Silva uses this. It also struggles to make sure all its input is transformed to unicode (among other ways using Formulator). In Plone, the situation is quite different -- its PlacelessTranslationService monkeypatches into the page template engine and puts in ways so that you can mix UTF-8 and unicode strings together. This then goes on to break assumptions of code that uses the page template engine in a unicode-pure environment (which is what happened to Silva). Regards, Martijn
Martijn Faassen wrote:
Jim Fulton wrote:
I forgot a very important need:
- Common approach to Unicode
In particular, In Zope 3, all text is stored and managed as Unicode. The publisher decodes request data and encodes response data. The vast majority of application and library code can ignore encoding issues. (The exceptions are applications and frameworks that need to exhange text with non-Unicode-aware external systems.) This has provided great simplifications and allowed us to avoid common pitfals from mixing Unicode and encoded text.
We need to migrate Zope 2 to use a similar strategy. We need volunteers to brainstorm how this can be done and make one or more proposals. This is likely a prerequisite for finishing the publisher and ZPT work.
This is definitely a scary topic, and I speak from years of experience with Zope 2 unicode here. This sounds like a very hard transition that would touch *a lot* of code in non-Zope 2 core. How do you envision all the form inputs to suddenly produce unicode strings, for instance?
We've struggled hard with Formulator to make it work with unicode for instance (and still it's buggy, as I wanted to support the non-unicode scenarios too). I can imagine any system in Zope that uses forms at all would need to be touched.
I'll volunteer to help brainstorm on this, but right now my brainstorm is only very dark and full of lightning.
You and I brainstormed this a few months ago. I think this was on the list. I think that, for starters, we would arrange that all Zope 3 views used in Zope 2 would get unicode input. If you like, I can try to find this discussion. :)
Anyway, in some basics, Zope 2 does have an approach to unicode for *output* that's fairly similar to Zope 3's: if you use unicode strings your entire output (including page templates) will be unicode (if you don't mix with non-unicode non-ascii strings..). Then the response encoding setting is read and everything is transformed once to unicode text. Silva uses this. It also struggles to make sure all its input is transformed to unicode (among other ways using Formulator).
In Plone, the situation is quite different -- its PlacelessTranslationService monkeypatches into the page template engine and puts in ways so that you can mix UTF-8 and unicode strings together. This then goes on to break assumptions of code that uses the page template engine in a unicode-pure environment (which is what happened to Silva).
Ick. I'm not suggesting this is easy. We may have some messy deprecation and backward compatibility code. But we *do* need to solve this problem eventually, and the solution doesn't get any closer without taking steps. Jim -- Jim Fulton mailto:jim@zope.com Python Powered! CTO (540) 361-1714 http://www.python.org Zope Corporation http://www.zope.com http://www.zope.org
Jim Fulton wrote:
Martijn Faassen wrote: [snip]
I'll volunteer to help brainstorm on this, but right now my brainstorm is only very dark and full of lightning.
You and I brainstormed this a few months ago. I think this was on the list. I think that, for starters, we would arrange that all Zope 3 views used in Zope 2 would get unicode input. If you like, I can try to find this discussion. :)
Ah, right, that is far less scary, indeed. Your post somehow gave me the impression you wanted to change the way current Zope 2 does things, but if you limit yourself to what happens with Zope 3 stuff in Zope 2, it's less scary. In fact Five already has hacks to make sure that unicode enters Five-generated forms. Replacing these hacks with something solid would be good.
Anyway, in some basics, Zope 2 does have an approach to unicode for *output* that's fairly similar to Zope 3's: if you use unicode strings your entire output (including page templates) will be unicode (if you don't mix with non-unicode non-ascii strings..). Then the response encoding setting is read and everything is transformed once to unicode text. Silva uses this. It also struggles to make sure all its input is transformed to unicode (among other ways using Formulator).
In Plone, the situation is quite different -- its PlacelessTranslationService monkeypatches into the page template engine and puts in ways so that you can mix UTF-8 and unicode strings together. This then goes on to break assumptions of code that uses the page template engine in a unicode-pure environment (which is what happened to Silva).
Ick.
I'm not suggesting this is easy. We may have some messy deprecation and backward compatibility code. But we *do* need to solve this problem eventually, and the solution doesn't get any closer without taking steps.
Yes. I'm optimistic about being able to do this for Five-related stuff. If this is eventually going to be people's main development system, then we can basically say we've solved the important unicode issues. What I'm worried about doing this for old code, but some steps will probably become clear during the brainstorming session. Migration tools that turn strings in the ZODB into unicode ones magically (with the ability to spell out exceptions and encoding)? Tricky... Regards, Martijn
Martijn Faassen wrote:
Jim Fulton wrote:
Martijn Faassen wrote:
[snip]
I'll volunteer to help brainstorm on this, but right now my brainstorm is only very dark and full of lightning.
You and I brainstormed this a few months ago. I think this was on the list. I think that, for starters, we would arrange that all Zope 3 views used in Zope 2 would get unicode input. If you like, I can try to find this discussion. :)
Ah, right, that is far less scary, indeed. Your post somehow gave me the impression you wanted to change the way current Zope 2 does things, but if you limit yourself to what happens with Zope 3 stuff in Zope 2, it's less scary.
In fact Five already has hacks to make sure that unicode enters Five-generated forms. Replacing these hacks with something solid would be good.
Anyway, in some basics, Zope 2 does have an approach to unicode for *output* that's fairly similar to Zope 3's: if you use unicode strings your entire output (including page templates) will be unicode (if you don't mix with non-unicode non-ascii strings..). Then the response encoding setting is read and everything is transformed once to unicode text. Silva uses this. It also struggles to make sure all its input is transformed to unicode (among other ways using Formulator).
In Plone, the situation is quite different -- its PlacelessTranslationService monkeypatches into the page template engine and puts in ways so that you can mix UTF-8 and unicode strings together. This then goes on to break assumptions of code that uses the page template engine in a unicode-pure environment (which is what happened to Silva).
Ick.
I'm not suggesting this is easy. We may have some messy deprecation and backward compatibility code. But we *do* need to solve this problem eventually, and the solution doesn't get any closer without taking steps.
Yes. I'm optimistic about being able to do this for Five-related stuff. If this is eventually going to be people's main development system, then we can basically say we've solved the important unicode issues.
What I'm worried about doing this for old code, but some steps will probably become clear during the brainstorming session. Migration tools that turn strings in the ZODB into unicode ones magically (with the ability to spell out exceptions and encoding)? Tricky...
I'd start by adding some deprecation warnings at the places where PlacelessTranslationService (or Localizer for that matter) do their workarounds so that normally incompatible types are mixed together. Florent -- Florent Guillaume, Nuxeo (Paris, France) CTO, Director of R&D +33 1 40 33 71 59 http://nuxeo.com fg@nuxeo.com
Florent Guillaume wrote:
Martijn Faassen wrote: [snip]
What I'm worried about doing this for old code, but some steps will probably become clear during the brainstorming session. Migration tools that turn strings in the ZODB into unicode ones magically (with the ability to spell out exceptions and encoding)? Tricky...
I'd start by adding some deprecation warnings at the places where PlacelessTranslationService (or Localizer for that matter) do their workarounds so that normally incompatible types are mixed together.
With the most recent Silva under development, we're not using either anymore, so someone else would have to do this. Where would you put the deprecation warning in PTS? It monkey patches the ZPT engine, and in particular its StringIO object, to make sure that UTF-8 strings and unicode strings, when finally concatenated at the end after page template generation, can actually be concatenated as it will first encode the unicode strings to UTF-8. The output of the page template will then be UTF-8. In Zope 2 without this in place, the output of the page template will be unicode, and if you put in UTF-8 strings (that aren't just ascii), the system will give rise to a UnicodeError when this concatenation takes place. I guess you could put some detection code in PTS's concatenation functionality to see whether anything unicode is mixed with non-ascii strings, and in that case output one deprecation warning. I have a suspicion that you'll see this warning whenever someone views a page template in Plone. :) Regards, Martijn
--On 9. Dezember 2005 10:11:42 -0500 Jim Fulton <jim@zope.com> wrote:
We need to migrate Zope 2 to use a similar strategy. We need volunteers to brainstorm how this can be done and make one or more proposals. This is likely a prerequisite for finishing the publisher and ZPT work.
My ZPT integration of the Z3 templates will definitely only allow unicode. Non-unicode content must pass either an encoding or accept that it will be converted using utf8 as default encoding to unicode. -aj
Andreas Jung wrote:
--On 9. Dezember 2005 10:11:42 -0500 Jim Fulton <jim@zope.com> wrote:
We need to migrate Zope 2 to use a similar strategy. We need volunteers to brainstorm how this can be done and make one or more proposals. This is likely a prerequisite for finishing the publisher and ZPT work.
My ZPT integration of the Z3 templates will definitely only allow unicode. Non-unicode content must pass either an encoding or accept that it will be converted using utf8 as default encoding to unicode.
Hm, so it auto-decodes non-unicode strings using UTF-8? That's a bit dangerous, as you suppress a large class of unicode errors in the code. Code that creates UTF-8 strings will be silently accepted. Not as bad as what PTS does, at least the output of the ZPT will be unicode, but scary nonetheless. Do you have any impression of how compatible is your code is with existing large Zope 2 codebases by the way? Regards, Martijn
--On 15. Dezember 2005 17:55:16 +0100 Martijn Faassen <faassen@infrae.com> wrote:
Hm, so it auto-decodes non-unicode strings using UTF-8? That's a bit dangerous, as you suppress a large class of unicode errors in the code. Code that creates UTF-8 strings will be silently accepted. Not as bad as what PTS does, at least the output of the ZPT will be unicode, but scary nonetheless.
Do you have any impression of how compatible is your code is with existing large Zope 2 codebases by the way?
I would like to keep the code as strict as possible for now. When the final implementation is ready we must test it with Plone and CPS and see what problems will occur and how we can deal with them in a sane way (hopefully saner than the current implementation). Otherwise we could also stick with the current implementation :-) So there is currently no need to hurry..I hope to finish the work in January...enough time left for testing. -aj
Hello. At Fri, 09 Dec 2005 10:11:42 -0500, Jim Fulton wrote:
- Common approach to Unicode
In particular, In Zope 3, all text is stored and managed as Unicode. The publisher decodes request data and encodes response data. The vast majority of application and library code can ignore encoding issues. (The exceptions are applications and frameworks that need to exhange text with non-Unicode-aware external systems.) This has provided great simplifications and allowed us to avoid common pitfals from mixing Unicode and encoded text.
We need to migrate Zope 2 to use a similar strategy. We need volunteers to brainstorm how this can be done and make one or more proposals. This is likely a prerequisite for finishing the publisher and ZPT work.
I want to tackle this problem, because I use japanese and I have experienced the encode/decode error pitfalls in a last few years. I'm wrong, I didn't touch zope development and didn't read the list carefully. In zope2.8, there is a zpt problem and this is a same pitfall. (unicode and string are mixed in TAL) -- TAHARA Yusei tahara@timedia.co.jp
--On 9. Dezember 2005 10:11:42 -0500 Jim Fulton <jim@zope.com> wrote:
I forgot a very important need:
- Common approach to Unicode
We need to migrate Zope 2 to use a similar strategy. We need volunteers to brainstorm how this can be done and make one or more proposals. This is likely a prerequisite for finishing the publisher and ZPT work.
I think there are two approaches. Textual content can be produced by nearly every Zope object (and its methods). Content can be composed by the basic Zope functionalities like DTML, ZPT, PyScripts and external methods (in the sense that these functionalities are able to call other objects and their methods). a) one could enforce all objects to return unicode (which would be a very hard requirement) and possibly break any application b) convert non-unicode content produced by Zope objects from where they are called (DTMl, ZPT, PyScript, Extmethods) to unicode. This would limit the number of places where we need to change code. The encoding of the non-unicode content could be from the 'content-type: XXX; charset=XXXX' header (if set) or as fallback from the configured zpublisher_default_encoding. An object could also set a property "my_output_encoding" (or so)... just-some-thoughts..., -aj
participants (5)
-
Andreas Jung -
Florent Guillaume -
Jim Fulton -
Martijn Faassen -
TAHARA Yusei