Re: [Jython-dev] PYTHONIOENCODING and

[Philip J. <pj...@un...>]

On Nov 30, 2013, at 4:08 PM, Jeff Allen wrote:

> I'm looking at a last test failure in test_file2k, which fails because 
> we do not support PYTHONIOENCODING (and for an incidental problem with 
> subprocess.Popen), but it asks more than a simple bug-fix or two. It 
> alerts us to a missing feature.
> 
> PYTHONIOENCODING was new in CPython 2.6, so 2.7 is the first version of 
> Jython that should support it. I think the implications are:
> 
> 1. Initialisation should read PYTHONIOENCODING to set the encoding 
> applicable to std(in|out|err) and the error policy applicable to 
> std(in|out). I can tell from the test this applies even if these streams 
> are not terminals.
> 
> 2. PYTHONIOENCODING is applied during PySystemState initialisation (so 
> it applies to the embedded interpreter) not just when using the command 
> org.python.util.jython. As a property of the stream it could not apply 
> to streams replaced with setIn, setOut, etc. e.g. under JSR-223 
> interpreters.
> 
> 3. We remove the -C <encoding> option from the Jython command line. 
> (It's not present in CPython 2.6 onwards: I can't find any documentation 
> for the CPython 2.5 command but assume it was superseded ... if it ever 
> existed. It's not documented for Jython 2.5 even.)

It's a really old Jython specific argument (8b2add647610). It looks like it was originally '-E' but I changed it to '-C' in 5b29613a1add for compatibility w/ CPython 2.5's -E. It didn't seem to bother anyone when it was changed. Maybe we should deprecate it, but since the -C change was a non event I'd be fine with removing it for 2.7.

> 
> 4. The Java property python.console.encoding if given would have 
> priority over PYTHONIOENCODING on console streams, that is where the 
> stream is a console according to isatty().

This property seems to just be the registry equivalent of -C (-E). So maybe we can kill it as well?

> 
> 5. The file type should have an errors property, and the errors and 
> encoding properties should by consulted when a PyUnicode is written, 
> whether to a pipe or the console. If it is a console, the console 
> handler will then decode (all) bytes written, using the same decoding to 
> present characters.
> 
> 6. file.encoding is not consulted when reading. If from a pipe, the pipe 
> supplies bytes and the read calls return them; if from the console, the 
> console handler has already encoded the keystrokes to bytes.
> 
> Does this sound about right?

LGTM

--
Philip Jenvey