I agree that supporting UTF-16 doesn't seem terribly useful. Also, thank you for giving the tokenizer some love! On Sat, Nov 14, 2015, at 11:19, Serhiy Storchaka wrote:

For now UTF-16 and UTF-32 source encodings are not supported. There is a comment in Parser/tokenizer.c:

/* Disable support for UTF-16 BOMs until a decision is made whether this needs to be supported. */

Can we make a decision whether this support will be added in foreseeable future (say in near 10 years), or no?

Removing commented out and related code will help to refactor the tokenizer, and that can help to fix some existing bugs (e.g. issue14811, issue18961, issue20115 and may be others). Current tokenizing code is too tangled.

If the support of UTF-16 and UTF-32 is planned, I'll take this to attention during refactoring. But in many places besides the tokenizer the ASCII compatible encoding of source files is expected.

_______________________________________________ Python-Dev mailing list Python-Dev@python.org https://mail.python.org/mailman/listinfo/python-dev Unsubscribe: https://mail.python.org/mailman/options/python-dev/benjamin%40python.org

On 11/14/2015 3:21 PM, Serhiy Storchaka wrote:

On 15.11.15 00:56, Victor Stinner wrote:

...

These encodings are rarely used. I don't think that any text editor use them. Editors use ascii, latin1, utf8 and... all locale encoding. But I don't know any OS using UTF-16 as a locale encoding. UTF-32 wastes disk space.

AFAIK the standard Windows editor Notepad uses UTF-16. And I often encountered Windows resource files in UTF-16. UTF-16 was more popular than UTF-8 on Windows some time. If this horse is dead I'll throw it away.

Just use UTF-8, ignoring an optional leading BOM. If someone wants to use something else, they can write a "preprocessor" to convert it to UTF-8 for use by Python.

The native encoding on Windows has been UTF-16 since Windows NT. Obviously we've survived without Python tokenization support for a long time, but every API uses it. I've hit a few cases where it would have been handy for Python to be able to detect it, though nothing I couldn't work around. Saying it is rarely used is rather exposing your own unawareness though - it could arguably be the most commonly used encoding (depending on how you define "used"). Cheers, Steve Top-posted from my Windows Phone -----Original Message----- From: "Victor Stinner" Sent: ‎11/‎14/‎2015 14:58 To: "Serhiy Storchaka" Cc: "python-dev@python.org" Subject: Re: [Python-Dev] Support of UTF-16 and UTF-32 source encodings These encodings are rarely used. I don't think that any text editor use them. Editors use ascii, latin1, utf8 and... all locale encoding. But I don't know any OS using UTF-16 as a locale encoding. UTF-32 wastes disk space. Ok, even if it exists, Python already accepts a very wide range of encoding. It is not worth to make the parser much more complex just to support encodings which are also never used (for .py files). Victor Le 14 nov. 2015 20:20, "Serhiy Storchaka" a écrit : For now UTF-16 and UTF-32 source encodings are not supported. There is a comment in Parser/tokenizer.c: /* Disable support for UTF-16 BOMs until a decision is made whether this needs to be supported. */ Can we make a decision whether this support will be added in foreseeable future (say in near 10 years), or no? Removing commented out and related code will help to refactor the tokenizer, and that can help to fix some existing bugs (e.g. issue14811, issue18961, issue20115 and may be others). Current tokenizing code is too tangled. If the support of UTF-16 and UTF-32 is planned, I'll take this to attention during refactoring. But in many places besides the tokenizer the ASCII compatible encoding of source files is expected. _______________________________________________ Python-Dev mailing list Python-Dev@python.org https://mail.python.org/mailman/listinfo/python-dev Unsubscribe: https://mail.python.org/mailman/options/python-dev/victor.stinner%40gmail.co...

On 11/14/2015 5:15 PM, Chris Angelico wrote:

Can the py.exe launcher handle a UTF-16 shebang? (I'm pretty sure Unix program loaders won't.) That alone might be a reason for strongly encouraging ASCII-compat encodings.

That raises an interesting question about if py.exe can handle a leading UTF-8 BOM. I have my emacs-on-Windows configured to store UTF-8 without BOM, but Notepad would put a BOM when saving UTF-8, last I checked.

On 11/14/2015 5:15 PM, Chris Angelico wrote:

I think even Notepad defaults to UTF-8 for files, now.

Just installed Windows 10 on a new machine, and upgraded to the latest Windows 10 release, 1511. Notepad defaults to ANSI encoding, as I think it always has. UTF-8 is an option, and it does seem to try to notice the original encoding of the file, when editing old files, but when creating a new one.... ANSI.

On 11/14/2015 5:37 PM, Chris Angelico wrote:

On Sun, Nov 15, 2015 at 12:27 PM, Glenn Linderman wrote:

...

Notepad defaults to ANSI encoding, as I think it always has. UTF-8 is an option, and it does seem to try to notice the original encoding of the file, when editing old files, but when creating a new one.... ANSI. Thanks. Is "ANSI" always an eight-bit ASCII-compatible encoding?

I wouldn't trust an answer to this question that didn't come from someone that used Windows with Chinese, Japanese, or Korean, as their default language for the install. So I don't have a trustworthy answer to give.

Glenn Linderman writes:

On 11/14/2015 5:37 PM, Chris Angelico wrote:

...

Thanks. Is "ANSI" always an eight-bit ASCII-compatible encoding?

I wouldn't trust an answer to this question that didn't come from someone that used Windows with Chinese, Japanese, or Korean, as their default language for the install. So I don't have a trustworthy answer to give.

AFAIK (I haven't actually used it as a default language, but I do know some details of their encodings) There are two main "issues" with the windows CJK encodings regarding ASCII compatibility: - There is a different symbol (a currency symbol) at 0x5c. Sort of. Most unicode translations of it will treat it as a backslash, and users do expect it to work for things like \n, path separators, etc, but it displays as ¥ or ₩. - Dual-byte characters can have ASCII bytes as their *second* byte.

On Sat, Nov 14, 2015 at 7:06 PM, Steve Dower wrote:

The native encoding on Windows has been UTF-16 since Windows NT. Obviously we've survived without Python tokenization support for a long time, but every API uses it.

Windows 2000 was the first version to have broad support for UTF-16. Windows NT (1993) was released before UTF-16, so its Unicode support is limited to UCS-2. (Note that console windows still restrict each character cell to a single WCHAR character. So a non-BMP character encoded as a UTF-16 surrogate pair always appears as two box glyphs. Of course you can copy and paste from the console to a UTF-16 aware window just fine.)

I've hit a few cases where it would have been handy for Python to be able to detect it, though nothing I couldn't work around.

Can you elaborate some example cases? I can see using UTF-16 for the REPL in the Windows console, but a hypothetical WinConIO class could simply transcode to and from UTF-8. Drekin's win-unicode-console package works like this.

Steve Dower writes:

Saying [UTF-16] is rarely used is rather exposing your own unawareness though - it could arguably be the most commonly used encoding (depending on how you define "used").

Because we're discussing the storage of .py files, the relevant definition is the one used by the Unicode Standard, of course: a text/plain stream intended to be manipulated by any conformant Unicode processor that claims to handle text/plain. File formats with in-band formatting codes and allowing embedded non-text content like Word, or operating system or stdlib APIs, don't count. Nor have I seen UTF-16 used in email or HTML since the unregretted days of Win2k betas[1] (but I don't frequent Windows- or Java-oriented sites, so I have to admit my experience is limited in a possibly relevant way). In Japan my impression is that modern versions of Windows have Memopad[sic] configured to emit UTF-8-with-signature by default for new files, and if not, the abomination known as Shift JIS (I'm not sure if that is a user or OEM option, though). Never a widechar encoding (after all, the whole point of Shift JIS was to use an 8-bit encoding for the katakana syllabary to save space or bandwidth). I think if anyone wants to use UTF-16 or UTF-32 for exchange of Python programs, they probably already know how to convert them to UTF-8. As somebody already suggested, this can be delegated to the py.exe launcher, if necessary, AFAICS. I don't see any good reason for allowing non-ASCII-compatible encodings in the reference CPython interpreter. However, having mentioned Windows and Java, I have to wonder about IronPython and Jython, respectively. Having never lived in either of those environments, I don't know what text encoding their users might prefer (or even occasionally encounter) in Python program source. Steve Footnotes: [1] The version of Outlook Express shipped with them would emit "HTML" mail with ASCII tags and UTF-8-encoded text (even if it was encodable in pure ASCII). No, it wasn't spam, either, so it probably really was Outlook Express as it claimed to be in one of the headers.

On 15 November 2015 at 07:23, Stephen J. Turnbull wrote:

I don't see any good reason for allowing non-ASCII-compatible encodings in the reference CPython interpreter.

From PEP 263:

Any encoding which allows processing the first two lines in the way indicated above is allowed as source code encoding, this includes ASCII compatible encodings as well as certain multi-byte encodings such as Shift_JIS. It does not include encodings which use two or more bytes for all characters like e.g. UTF-16. The reason for this is to keep the encoding detection algorithm in the tokenizer simple. So this pretty much confirms that double-byte encodings are not valid for Python source files. Paul

I'm approaching this from the premise that we would like to avoid needless surprises for users not versed in text encoding. I did a simple experiment with notepad on Windows 7 as if a naïve user. If I write the one-line program: print("Hello world.") # by Jeff It runs, no surprise. We may legitimately encounter Unicode in string literals and comments. If I write: print("j't'kif Anaïs!") # par Hervé and try to save it, notepad tells me this file "contains characters in Unicode format which will be lost if you save this as an ANSI encoded text file." To keep the Unicode information I should cancel and choose a Unicode option. In the "Save as" dialogue the default encoding is ANSI. The second option "Unicode" is clearly right as the warning said "Unicode" 3 times and I don't know what big-endian or UTF-8 mean. Good that worked. Closed and opened it looks exactly as I typed it. But the bytes I actually wrote on disk consist of a BOM and UTF-16-LE. And running it I get: File "bonjour.py", line 1 SyntaxError: Non-UTF-8 code starting with '\xff' in file bonjour.py on line 1, but no encoding declared; see http://python.org/dev/peps/pep-0263/ for details If I take the hint here and save as UTF-8, then it works, including printing the accent. Inspection of the bytes shows it starts with a UTF-8 BOM. In Jython I get the same results (choking on UTF-16), but saved as UTF-8, it works. I just have to make sure that's a Unicode constant if I want it to print correctly, as we're at 2.7. Jython has a checkered past with encodings, but tries to do exactly the same as CPython 2.7.x. Now, a fact I haven't mentioned is that my machine was localised to simplified Chinese (to diagnose some bug) during this test. If I re-localise to my usual English (UK), I do not get the guidance from notepad: instead it quietly saves as Latin-1 (cp1252), perhaps because I'm in Western Europe. Python baulks at this, at the first accented character. If I save from notepad as Unicode or UTF-8 the results are as before, including the BOM. In some circumstances, then, the natural result of using notepad and not sticking to ASCII may be UTF-16-LE with a BOM, or Latin-1 depending on localisation, it seems. The Python error message provides a clue what a user should do, but they would need some background, a helpful teacher, or the Internet to sort it out. Jeff Allen On 15/11/2015 07:23, Stephen J. Turnbull wrote:

Steve Dower writes:

...

Saying [UTF-16] is rarely used is rather exposing your own unawareness though - it could arguably be the most commonly used encoding (depending on how you define "used").

Because we're discussing the storage of .py files, the relevant definition is the one used by the Unicode Standard, of course: a text/plain stream intended to be manipulated by any conformant Unicode processor that claims to handle text/plain. File formats with in-band formatting codes and allowing embedded non-text content like Word, or operating system or stdlib APIs, don't count. Nor have I seen UTF-16 used in email or HTML since the unregretted days of Win2k betas[1] (but I don't frequent Windows- or Java-oriented sites, so I have to admit my experience is limited in a possibly relevant way).

In Japan my impression is that modern versions of Windows have Memopad[sic] configured to emit UTF-8-with-signature by default for new files, and if not, the abomination known as Shift JIS (I'm not sure if that is a user or OEM option, though). Never a widechar encoding (after all, the whole point of Shift JIS was to use an 8-bit encoding for the katakana syllabary to save space or bandwidth).

I think if anyone wants to use UTF-16 or UTF-32 for exchange of Python programs, they probably already know how to convert them to UTF-8. As somebody already suggested, this can be delegated to the py.exe launcher, if necessary, AFAICS.

I don't see any good reason for allowing non-ASCII-compatible encodings in the reference CPython interpreter.

However, having mentioned Windows and Java, I have to wonder about IronPython and Jython, respectively. Having never lived in either of those environments, I don't know what text encoding their users might prefer (or even occasionally encounter) in Python program source.

Steve

Footnotes: [1] The version of Outlook Express shipped with them would emit "HTML" mail with ASCII tags and UTF-8-encoded text (even if it was encodable in pure ASCII). No, it wasn't spam, either, so it probably really was Outlook Express as it claimed to be in one of the headers.

_______________________________________________ Python-Dev mailing list Python-Dev@python.org https://mail.python.org/mailman/listinfo/python-dev Unsubscribe:https://mail.python.org/mailman/options/python-dev/ja.py%40farowl.co.uk

Random832 writes:

"Stephen J. Turnbull" writes:

...

I don't see any good reason for allowing non-ASCII-compatible encodings in the reference CPython interpreter.

There might be a case for having the tokenizer not care about encodings at all and just operate on a stream of unicode characters provided by a different layer.

That's exactly what the PEP 263 implementation does in Python 2 (with the caveat that Python 2 doesn't know anything about Unicode, it's a UTF-8 stream and the non-ASCII characters are treated as bytes of unknown semantics, so they can't be used in syntax). I don't know about Python 3, I haven't looked at the decoding of source programs. But I would assume it implements PEP 263 still, except that since str is now either widechars or PEP 393 encoding (ie, flexible widechars) that encoding is now used instead of UTF-8. I'm sure that there are plenty of ASCII-isms in the tokenizer in the sense that it assumes the ASCII *character* (not byte) repertoire. But I'm not sure why Serhiy thinks that the tokenizer cares about the representation on-disk. But as I say, I haven't looked at the code so he might be right. Steve

On 15 November 2015 at 16:40, Stephen J. Turnbull wrote:

What PEP 263 did do was to specify that non-ASCII-compatible encodings are not supported by the PEP 263 mechanism for declaring the encoding of a Python source program. That's because it looks for a "magic number" which is the ASCII-encoded form of "coding:" in the first two lines. It doesn't rule out alternative mechanisms for encoding detection (specifically, use of the UTF-16 "BOM" signature); it just doesn't propose implementing them.

That was my initial thought. But combine this with the statement from the language docs that the default encoding when there is no PEP 263 encoding specification is UTF-8 (or ASCII in Python 2) and there's no valid way that I can see that a UTF-16 encoding could be valid (short of a formal language change). Anyway, Guido has spoken, so I'll leave it there. Paul

On 14.11.2015 23:56, Victor Stinner wrote:

These encodings are rarely used. I don't think that any text editor use them. Editors use ascii, latin1, utf8 and... all locale encoding. But I don't know any OS using UTF-16 as a locale encoding. UTF-32 wastes disk space.

UTF-16 is used a lot for Windows text files, e.g. Unicode CSV files (the save as "Unicode text file" option writes UTF-16). However, nowadays, all text editors also support UTF-8 and many of these recognize the UTF-8 BOM as identifier to detect Unicode text files.

Ok, even if it exists, Python already accepts a very wide range of encoding. It is not worth to make the parser much more complex just to support encodings which are also never used (for .py files).

Agreed. In Python 2 we decided to only allow ASCII super-sets for Python source files, which out ruled multi-byte encodings such as UTF-16 and -32. I don't think we need to make the parser more complex just to support them. UTF-8 works fine as Python source code encoding.

Victor Le 14 nov. 2015 20:20, "Serhiy Storchaka" a écrit :

...

For now UTF-16 and UTF-32 source encodings are not supported. There is a comment in Parser/tokenizer.c:

/* Disable support for UTF-16 BOMs until a decision is made whether this needs to be supported. */

Can we make a decision whether this support will be added in foreseeable future (say in near 10 years), or no?

Removing commented out and related code will help to refactor the tokenizer, and that can help to fix some existing bugs (e.g. issue14811, issue18961, issue20115 and may be others). Current tokenizing code is too tangled.

If the support of UTF-16 and UTF-32 is planned, I'll take this to attention during refactoring. But in many places besides the tokenizer the ASCII compatible encoding of source files is expected.

_______________________________________________ Python-Dev mailing list Python-Dev@python.org https://mail.python.org/mailman/listinfo/python-dev Unsubscribe: https://mail.python.org/mailman/options/python-dev/victor.stinner%40gmail.co...

_______________________________________________ Python-Dev mailing list Python-Dev@python.org https://mail.python.org/mailman/listinfo/python-dev Unsubscribe: https://mail.python.org/mailman/options/python-dev/mal%40egenix.com

-- Marc-Andre Lemburg eGenix.com Professional Python Services directly from the Experts (#1, Nov 15 2015)

...

...

Python Projects, Coaching and Consulting ... http://www.egenix.com/ Python Database Interfaces ... http://products.egenix.com/ Plone/Zope Database Interfaces ... http://zope.egenix.com/

2015-10-23: Released mxODBC Connect 2.1.5 ... http://egenix.com/go85 ::: We implement business ideas - efficiently in both time and costs ::: eGenix.com Software, Skills and Services GmbH Pastor-Loeh-Str.48 D-40764 Langenfeld, Germany. CEO Dipl.-Math. Marc-Andre Lemburg Registered at Amtsgericht Duesseldorf: HRB 46611 http://www.egenix.com/company/contact/ http://www.malemburg.com/