ERA001: autofix removes valid comments that do not contain code or removes not everything

[saippuakauppias]

Example:

# Example of headers:
# X-token: <token_id>
# X-scope: no-scope
# End of comment about headers


# {
#    "some-key": {
#         '1': 1,
#         '2': 2',
#     }
# }


# some = 1
# awesome = 2
#
# foobar = 3
# barfoo = 4


# example code from IPython:
# In [20]: parse.parse_qsl('&l=55.123;123.1&count=3', keep_blank_values=1)
# Out[20]: [('l', '55.123'), ('123.1', ''), ('count', '3')]
# End of comment


# (пример, пример2) = (example1, example2)


# if not success:  # TODO: comment
#    errors.append('Ошибка: error')


# comment
# 1. [{пример первый}]
# 2. [[пример]]
# 3. [] - пусто


# Пример: {"description": "weight: "}


# comment about calculation
# 100 / 2 = 50
# end of comment


# comment:
# - tuple: ()
# - list: []
# - set: set()
# - dict: {}


# example of output:
# 'OK: message sent. Message-ID: 123456'


# MySQL: enum('positive', 'negative')


# query: lock + user.id

Ruff output:

$ ruff --fix --select ERA001 ruff_errors/ERA001.py
Found 24 errors (24 fixed, 0 remaining).

Contents of the file after fix:

# Example of headers:
# X-token: <token_id>
# End of comment about headers


#    "some-key": {


#


# example code from IPython:
# End of comment




# if not success:  # TODO: comment


# comment
# 1. [{пример первый}]
# 3. [] - пусто




# comment about calculation
# end of comment


# comment:


# example of output:

[charliermarsh]

I don't think we're in a position to improve this much, the logic itself is derived from eradicate, and I'd bet eradicate has the same or very similar behavior. Perhaps we should remove the autofix from this rule though. It seems correct to flag that there's some commented-out code here, but rely on users to remove it.

Alternatively, we could only autofix if all lines in a comment block are identified as code.

[dhruvmanila]

We could make this as a Suggested (Applicability::Suggested) fix.

    /// The fix may be what the user intended, but it is uncertain.
    /// The fix should result in valid code if it is applied.
    /// The fix can be applied with user opt-in.

[zanieb]

I like

Alternatively, we could only autofix if all lines in a comment block are identified as code.

Perhaps we could use Suggested in that case and Manual otherwise?

[karolinepauls]

Another example: it thinks the comment is code even though params isn't a valid keyword

# params: abc
query = """
SELECT :abc;
"""

[dhruvmanila]

Another example: it thinks the comment is code even though params isn't a valid keyword

# params: abc
query = """
SELECT :abc;
"""

I believe this is considered as a valid code (variable: annotation):

params: int

[bodograumann]

I am absolutely in favor of not automatically removing these comments in any situation. Yes, commented-out code is usually a mistake, but you really don't want to loose actual comments accidentally.

Our example:

# foo and bar (JIRA-9999)

[zanieb]

@bodograumann what is the benefit of enabling the ERA001 rule for you then?

[bodograumann]

It will still warn me and usually correctly so, but I still need to decide myself whether a line needs to be removed.

[zanieb]

We've updated the applicability of the fix here to "Manual" which will never automatically apply once we merge #5119 — I don't think there's more to do here this issue will be resolved once applicability is respected by the CLI per #4185

[jaap3]

Was about to open a new issue, but I think this fits here:

While testing out ruff, I ran into some false positives for ERA001 that confused me a bit.

Some examples:

# Keep this list sorted alphabetically
# (it is used in example)

# Check if the ip address in either:
# local (loopback) or internal (private)

##
# Client (consumers)
##

...

##
# Server (protocol)
##

All trigger ERA001 even though they are not really code.

I had also hoped that code fragments that are part of a larger comment wouldn't trigger ERA001, yet all of these do:

# XXX: Cannot use:
# self.example(foo, 'bar')
# because of issue #1337

# XXX: Example category has a duplicate 'code' and is omitted
# EXAMPLE_CATEGORY = '1.0', 'Example Category'

# Note: Admin site instance should be passed when calling
# ExampleAdminView.as_view(site=site)

# main categories themselves are not part of the return value of
# get_subcategories()

This is all using ruff 0.0.286.

I've already marked ERA001 as unfixable in my configuration file. I'm hoping it is possible to tweak the heuristics of ERA001 a bit so it takes the surrounding context into account or something?

For now I'll just be working around this by rewording/reformatting these comments until they don't trigger anymore.

[zanieb]

Yeah the ERA001 heuristics are very prone to false positives, unfortunately. I think there's room for improvement, but it's not entirely clear how.

def foo():
    x = 1
    # add to x
    # x += 1
   return x

I think this should raise as an ERA001 violation but if we ignore comments that are part of a larger non-code context then we won't raise here. Maybe that's worth the false negative to reduce the false positives.

[jaap3]

This seems like a bug:

A single parenthesis triggers ERA001:

echo "# )" | ruff --select ERA001,RUF100 --stdin-filename test.py
test.py:1:1: ERA001 [*] Found commented-out code

Adding a noqa to that line triggers RUF100:

echo "# )  # noqa: ERA001" | ruff --select ERA001,RUF100 --stdin-filename test.py
test.py:1:6: RUF100 [*] Unused `noqa` directive (unused: `ERA001`)

[charliermarsh]

@jaap3 - Funnily enough that specific case was reported recently and is fixed on main.

[tusharsadhwani]

Is there scope for a PR for better algorithms, and a better set of false positive rules here?

It will deviate from the eradicate plugin, would that be okay or should it be a different rule entirely then?

[charliermarsh]

Yeah definitely fine to fold in improvements even if it deviates from eradicate.

[tusharsadhwani]

Self-documenting here for later.

Examples of false negatives:

# if foo:

# elif bar:

# if (
#  x
# ):

# else:

Examples of false positives:

# continue
# pragma: cover
# (something)
# O(1)
# this is O(1)