Fix incorrect backslash before multi-lines string

[LiuYinCarl]

The PR is used to fix incorrect backslash before multi-lines string which will result in incorrect lines alignment.

Source code for test

#################### case 1
a = ["aaa",\
     "bbb \
     ccc"]

#################### case 2
b = ("a",\
     "b \
     c")

########## case 3
c = "xxx"\
    "yyy \
    zzz"

#################### case 4
def f(a, b):
    pass

f(100,\
  "hello \
  world")

#################### case 5(good case)
d = """\
hello
"""


def show_incorrect_line_align():
    # line 1
    # line 2
    # line 3
    # line 4
    # line 5
    pass

Run commands

coverage run main.py
coverage html
cd htmlcov
python3 -m http.server 9000

[LiuYinCarl]

@nedbat Hi,do you have time to take a kook at this PR? Thanks.

[nedbat]

I will take a look, but it might take me some time.

[nedbat]

BTW, what name should I use for the Changelog and Contributors files?

[nedbat]

Tests added in dc819ff
Changelog updated in da1682f
Thanks!

[nedbat]

This is now released as part of coverage 7.6.1.

[the-13th-letter]

Tests added in dc819ff

I stumbled across this while reviewing #1838. I think the tests (and this fix) are incomplete.

coveragepy/tests/test_html.py

Lines 1134 to 1154 in dc819ff

	def test_bug_1828(self) -> None:
	# https://github.com/nedbat/coveragepy/pull/1828
	self.make_file("backslashes.py", """\
	a = ["aaa",\\
	"bbb \\
	ccc"]
	""")
	cov = coverage.Coverage()
	backslashes = self.start_import_stop(cov, "backslashes")
	cov.html_report(backslashes, directory="out")
	contains(
	"out/backslashes_py.html",
	# line 2 is `"bbb \`
	r'<a id="t2" href="#t2">2</a></span>'
	+ r'<span class="t"> <span class="str">"bbb \</span>',
	# line 3 is `ccc"]`
	r'<a id="t3" href="#t3">3</a></span>'
	+ r'<span class="t"><span class="str"> ccc"</span><span class="op">]</span>',
	)

In lines 1146–1154, you are testing whether the HTML output correctly reproduces the backslash in the "bbb ccc" string. However, the rendered text does not contain the backslash and linebreak within line 1137. This leads to a missing linebreak, and therefore to misaligned coverage information similar to #1836, just in the other direction.

coveragepy/tests/test_phystokens.py

Lines 101 to 117 in dc819ff

	def test_1828(self) -> None:
	# https://github.com/nedbat/coveragepy/pull/1828
	tokens = list(source_token_lines(textwrap.dedent("""
	x = \
	1
	a = ["aaa",\\
	"bbb \\
	ccc"]
	""")))
	assert tokens == [
	[],
	[('nam', 'x'), ('ws', ' '), ('op', '='), ('ws', ' '), ('num', '1')],
	[('nam', 'a'), ('ws', ' '), ('op', '='), ('ws', ' '),
	('op', '['), ('str', '"aaa"'), ('op', ','), ('xx', '\\')],
	[('ws', ' '), ('str', '"bbb \\')],
	[('str', ' ccc"'), ('op', ']')],
	]

Testing the same input, just with the tokenizer instead of the HTML report renderer. You can see in line 112 that there is a ws token with many spaces (which doesn't occur in the tokenizer input) instead of two tokens with a backspace continuation (which does occur). Again, this leads to a missing linebreak, and to misaligned coverage info for all following lines.

---

Concerning this pull request in general:

• Python 3.11 (Linux) with coverage 7.6.1 (plus commit 01779db from fix backslash in f-string #1838 applied manually): I see the "before fix"/incorrect version for the a variable and the "after fix"/correct version for the b, c, d and f variables. The same happens whether I use single-quoted strings, triple-quoted strings, or triple-quoted f-strings.
• Python 3.12 (Linux) with coverage 7.6.1 (plus commit 01779db): I see the incorrect version for the a variable if and only if I'm not using triple-quoted f-strings, and the correct version in all other cases/for all other variables.

[the-13th-letter]

I stumbled across this while reviewing #1838. I think the tests (and this fix) are incomplete.

My apologies, I forgot to ping @nedbat, as the author of these tests.

See also #1838 (comment) for a sample file-plus-coverage-output of a run which is correct according to (my reading of) the aforementioned test functions, despite misaligning coverage info later on in the file. Screenshot:

[nedbat]

I appreciate all the work you are doing. I'm trying to keep up! You said:

In lines 1146–1154, you are testing whether the HTML output correctly reproduces the backslash in the "bbb ccc" string. However, the rendered text does not contain the backslash and linebreak within line 1137.

When I print the HTML created by this test, I see:

<main id="source">
    <p class="run"><span class="n"><a id="t1" href="#t1">1</a></span><span class="t"><span class="nam">a</span> <span class="op">=</span> <span class="op">[</span><span class="str">"aaa"</span><span class="op">,</span><span class="xx">\</span>&nbsp;</span><span class="r"></span></p>
    <p class="pln"><span class="n"><a id="t2" href="#t2">2</a></span><span class="t">     <span class="str">"bbb \</span>&nbsp;</span><span class="r"></span></p>
    <p class="pln"><span class="n"><a id="t3" href="#t3">3</a></span><span class="t"><span class="str">     ccc"</span><span class="op">]</span>&nbsp;</span><span class="r"></span></p>
</main>

Line 1 ends with backslash, and is separate from line 2. Can you show me more specifically what you mean?

[LiuYinCarl]

@nedbat maybe it's because the code of @the-13th-letter contain a blank between "," and "\", I fix it in a1aa6b8

[nedbat]

To help with these kinds of tests, I added a new helper to extract the rendered test from an HTML report in 9494ba7.

[the-13th-letter]

@netbat:

When I print the HTML created by this test, I see:

<main id="source">
    <p class="run"><span class="n"><a id="t1" href="#t1">1</a></span><span class="t"><span class="nam">a</span> <span class="op">=</span> <span class="op">[</span><span class="str">"aaa"</span><span class="op">,</span><span class="xx">\</span>&nbsp;</span><span class="r"></span></p>
    <p class="pln"><span class="n"><a id="t2" href="#t2">2</a></span><span class="t">     <span class="str">"bbb \</span>&nbsp;</span><span class="r"></span></p>
    <p class="pln"><span class="n"><a id="t3" href="#t3">3</a></span><span class="t"><span class="str">     ccc"</span><span class="op">]</span>&nbsp;</span><span class="r"></span></p>
</main>

Line 1 ends with backslash, and is separate from line 2. Can you show me more specifically what you mean?

Currently, the test_bug_1828 function does not inspect or assert anything about the first line of the output; it only checks that the "bbb \<newline>ccc" part of the output is as expected. So both your rendition and my rendition would pass this test, even though my rendition has the "missing linebreak/coverage info misalignment" issue. That's why I'm arguing that the tests are incomplete, because methinks they should be explicitly asserting the comma-(space-)backslash in the first line too.

---

I'm going to be busy at $WORK for the next couple of days, and will only be able to return to debugging this issue on Saturday (Aug 31st).

@nedbat maybe it's because the code of @the-13th-letter contain a blank between "," and "\", I fix it in a1aa6b8

Sounds plausible. I'll try that out too on Saturday.

[nedbat]

Currently, the test_bug_1828 function does not inspect or assert anything about the first line of the output;

I changed that this morning: now it makes an assertion about all of the rendered text. I hope we can write more tests using the same technique.

[the-13th-letter]

Currently, the test_bug_1828 function does not inspect or assert anything about the first line of the output;

I changed that this morning: now it makes an assertion about all of the rendered text. I hope we can write more tests using the same technique.

Wonderful. I've taken the liberty of extending test_bug_1828 to cover more cases of whitespace-then-continuation-backslash. On current HEAD (28d22a3), I cannot get this test to break anymore, and it checks all the rendered text(-contents) against our expected output. Please see the attached patch, which passes all linting and all pytest tests (though perhaps not all code style checks) on Python 3.11 and 3.12.

(Installing 3.8, 3.9, 3.10 or PyPy on my system – with development headers – is quite a pain, so I only checked 3.11 and 3.12.)

(Yes, I know, GitHub has proper pull requests, but this would be a pull request on a pull request, and it feels a little overkill for something like this.)
issue_1828_more_whitespace_tests.diff.txt