[pypy-dev] RPython unicode support design question

[Armin Rigo]

Hi Carl,

On Tue, Nov 06, 2007 at 01:07:29PM +0100, Carl Friedrich Bolz wrote:
> My proposal is the following: forbid _all_ mixing of strings and unicode 
> in RPython.

Right now we allow mixing plain and unicode characters, at least in the
annotator.  There is at least one use case for this: you can write in a
uniform way some code that gets two specialized versions, one for string
and one for unicode.  I use this e.g. in objspace.std.formatting, where
two versions of the formatter build a list called 'result' in both
cases, but which is either a list of chars or a list of unichars.  It's
full of code like

    result += '0' * padding

which works in both cases because the character is turned into a unichar
if necessary - that is, if 'result' is already known to be a list of
unichars.

So while I see the motivation for forbidding implicit conversion, I also
think that there are good use cases where it makes sense to have it - at
least for single characters.  I guess that, as usual, the "RPython way"
is to forbid implicit conversions from full strings to full unicode
strings, until the day where we have a good use case for it :-)


Armin