lark
lark copied to clipboard
lark-parser
Breaking changes / docs out of date
Describe the bug
After updated from Lark 1.1.5 to 1.1.7, the parser fails with (at least) two problems. These seem to be due to a rename of ParseTree to Tree (contrary to the docs e.g. on parse: https://lark-parser.readthedocs.io/en/stable/classes.html#lark.Lark.parse) and a re-ordering of v_args in the Transformer methods to no longer match what's documented here:
https://lark-parser.readthedocs.io/en/stable/visitors.html#v-args
Also, the docs don't seem to have been building for the past few weeks:
https://readthedocs.org/projects/lark-parser/builds/
To Reproduce
Write a transformer with something like this:
@v_args(meta=True)
def term_ref(self, meta: Meta, items: List[Any]) -> Any:
s = str(items[0])
and it will fail by saying Meta object not subscriptable. This works fine in 1.1.5, but it looks like 1.1.7 expects this instead:
@v_args(meta=True)
def term_ref(self, items: List[Any], meta: Meta) -> Any:
s = str(items[0])
Considering that v_args(meta=True) is being tested with the order meta, items, and those tests still pass, I don't think that changed:
https://github.com/lark-parser/lark/blob/ba5ae311569b3ede4f712808e6dfdfe5162461c2/tests/test_trees.py#L209
Not sure what you mean with the rename?
Are you sure you are actually looking at 1.1.7? Both of these changes are stuff from pre-1.0.0
Having uninstalled 1.1.5 and installed 1.1.7 again, I can't reproduce the argument order problem, so maybe something funny happened with my pip version locally. If that's the case, the only problems seem to be the reference to ParseTree here: https://lark-parser.readthedocs.io/en/stable/classes.html#lark.Lark.parse (n.b. this is listed as the return type of parse, but it seems to be called Tree now) and the fact that the docs aren't building on ReadTheDocs.
Ah! I found it. I had this (I guess defunct?) package installed in my global site_packages:
https://pypi.org/project/lark-parser/
It's a bit confusing that the readthedocs project is called lark-parser but on pypi its lark. I guess this is because lark was already taken. I don't know if it's possible/desirable to take the pypi lark-parser offline, or at least warn people not to use it.
I don't want to take lark-parser offline from pypi, because it might break existing projects that use an old version. However, we don't publish any new packages there. What sort of warning do you think I should place, and where?
Yes, https://lark.readthedocs.io/ is occupied by an unmaintained project.
@MegaIng IIRC he willingly relinquished lark on PyPI. Perhaps we can ask him to do the same with readthedocs?
Perhaps you could publish a final version under lark-parser that either issues a deprecation warning whenever the module is imported or is just a stub depending on the latest version of lark. There seem to be some reasonable suggestions in this article:
https://www.dampfkraft.com/code/how-to-deprecate-a-pypi-package.html
People should be able to keep the old behaviour if they depend on lark-parser==0.12.0.