Support for case-insensitive matching of literals?

[kirberich]

I apologise if I missed this somewhere, but I have a grammar with a lot of string literals that are case insensitive, and I cannot find anything in the docs about how to parse this best, here's one example:

unary_operator <- ~("abs" / "acos" / "asin" / "atan" / "cos" / "exp" / "fix" / "fup" / "ln" / "round" / "sin" / "sqrt" / "tan")

The best I've been able to do is to replace every literal with a combination of class matches, e.g. [aA][bB][sS] instead of "abs", which I guess works but does become tedious, so I just wanted to check if there's something I'm missing!

[goodmami]

Nope, you didn't miss anything. This is not currently a feature of Pe.

I agree that writing out everything with sequences of character classes is tedious and I'm not opposed to adding such a feature, but there's enough complexity in the issue that I'd like to have a plan or something easy to reverse.

As for the backend implementation, Python's str.casefold() method is preferred for Unicode caseless matching, but it doesn't handle locale-specific things like the Turkic dotless-i (ı) with the same uppercase as the dotted-i (I):

>>> 'ı' == "I".lower()  # I -> i
False
>>> 'ı'.upper() == "I"  # ı -> I
True
>>> 'ı'.casefold() == "I".casefold()  # ı != i
False

For this case and some others, Python's re module has a list of special cases: https://github.com/python/cpython/blob/main/Lib/re/_casefix.py

>>> import re
>>> re.match("ı", "I", flags=re.IGNORECASE)
<re.Match object; span=(0, 1), match='I'>
>>> re.match("I", "ı", flags=re.IGNORECASE)
<re.Match object; span=(0, 1), match='ı'>

As for how this would be usable in Pe, functions like pe.compile() have a flags parameter for more global things (but, to be honest, I'm not very happy with how this currently works and may change it in the future), but not for things at a rule- or term-level. We could introduce some new metasyntax similar to re's (?i: ... ) (maybe (...)^i), or a parameter like RULE_FLAGS={"unary_operator": pe.IGNORECASE}.

[kirberich]

That makes a lot of sense, thanks for the detailed answer! I realised that in my case I could actually also get away with just accepting [a-zA-Z] and then checking for supported operations in the action later if it all became too much, but of course that's not generally true, and it does make the grammar less informative.