On Thu, Nov 7, 2019 at 8:47 AM Greg Ewing firstname.lastname@example.org wrote:
Andrew Barnert via Python-ideas wrote:
On Nov 6, 2019, at 08:59, Chris Angelico email@example.com wrote:
No, because "x <-- y" is already legal syntax
You could handle that by making the grammar more complicated.
Or just have the tokeniser treat "<--" as a single token, the same way that it treats "<=" as a single token rather than "<" followed by "=". It would be a backwards-incompatible change (if you really wanted "less than minus minus something" you'd have to put a space in somewhere) but replacing the assignment operator is already a much bigger one.
To clarify: I wasn't saying that it's fundamentally impossible to have these kinds of parsing rules, but that it's backward incompatible. Notably, even though this syntax is fairly unlikely to come up, it means that anyone using "<--" as an assignment operator will have to worry about older Python versions misinterpreting it. If you create a brand new operator out of something that's currently invalid syntax, then it's easy - you get an instant compilation error on an older interpreter. With this, it might sometimes result in a runtime NameError or TypeError, and even worse, might just silently do the wrong thing. That's why Python 3.9 still won't let you write "except ValueError, IndexError:" - you *have* to parenthesize the tuple, because the comma syntax had a different meaning in Python 2 (the "except Exception as name:" syntax was backported to 2.6/2.7 but the older syntax is of course still valid). There is no way that you can accidentally run your code on the wrong Python and have it silently assign to IndexError instead of catching two types.