Black: Black 19.10b0 no longer fixes short argument lists to a single line

[jdufresne]

Describe the bug

In previous versions of Black, short argument lists would be wrapped to a single line. In new versions of Black, this is not always the case.

If the argument list starts off as a single line, it'll remain a single line. If the argument list starts as multiple lines, it'll remain multiple lines. See test case below.

To Reproduce

test.py

my_func(
    "a",
    "b",
)

With Black 19.3b0

@@ -1,5 +1,2 @@
-my_func(
-    "a",
-    "b",
-)
+my_func("a", "b")

With Black 19.10b0

@@ -1,5 +1,4 @@
 my_func(
-    "a",
-    "b",
+    "a", "b",
 )

Expected behavior

I expect the behavior experienced in 19.3b0.

Environment

• Version: 19.10b0 (also master)
• OS and Python version: Fedora 30 Python 3.7.5

Does this bug also happen on master?

Yes.

https://black.now.sh/?version=master&state=_Td6WFoAAATm1rRGAgAhARYAAAB0L-Wj4ABjAEZdAD2IimZxl1N_WmBpTZzyXdYgGAumRPJ9fzc8wrFrsw5q41oEoZU7ytbygjdQqX-b0UdsSVEi1fbjjwWIr_X0wjnhZyweKwAAAABPNX1pNP1yrQABYmRPGZI9H7bzfQEAAAAABFla

[jdufresne]

Perhaps related, but I noticed that Black may no longer remove a trailing comma inside a single line argument list:

https://black.now.sh/?version=master&state=_Td6WFoAAATm1rRGAgAhARYAAAB0L-Wj4ABXAENdAD2IimZxl1N_WlkOE2WaAGC0rYwFwepCEK6-2b1ZbHEIRIItZMyY8a56ieNAhog7VJVt-X0IF4ViKQXsyGu6qISxtncAADZ5kboJyk78AAFfWHYtFXgftvN9AQAAAAAEWVo=

[jdufresne]

Both bisected to commit 9854d4b375ef84d651892785165a6c7b0e9f391b.

CC @durin42

[jdufresne]

Perhaps another version of #1129. They bisect to the same commit.

[durin42]

This is WAI: as part of #826 we have to preserve trailing commas. It's a little surprising to me that it reformats that argument list /at all/, but I do recall something is "funny" about function calls from when I wrote the patch.

[jdufresne]

FWIW, as an end user, I find this new beahvior quite suprising and a move away from the opinionated philosophy I really love about Black. When I write Python code, I almost never think about formatting anymore because I know Black will take of it later. With this change, it means I'm back to thinking about formatting. "Should I add a trailing comma or remove it?", is now a question I'll be asking before submitting code.

I normally blindly add a trailing comma to all code. Not because I want to keep arguments across multiple lines, but just out of habit. I don't think about it much more than that and let Black figure it out later.

The PR that introduce this change is all about preserving multiple lines in collections. So what does that have to do with argument lists of function calls? Are you intentionally grouping argument lists in with collections here, or is that an unintended side effect of the implementation?

I think there is another problem with this change. Consider a very long argument list:

my_func("abcabcabc123123", "abcabcabc123123", "abcabcabc123123", "abcabcabc123123", "abcabcabc123123")

After formatting using master, this becomes (notice a trailing comma was added):

my_func(
    "abcabcabc123123",
    "abcabcabc123123",
    "abcabcabc123123",
    "abcabcabc123123",
    "abcabcabc123123",
)

Now -- perhaps due to refactoring -- the last 2 arguments are removed and reformatted:

my_func(
    "abcabcabc123123", "abcabcabc123123", "abcabcabc123123",
)

Now, because of this new change and the added trailing comma, Black doesn't reformat the line back to a single line.

[vemel]

Trailing comma should be handled differently for LHS and RHS. For LHS it can be safely removed from a function definition. For RHS it should always be preserved.

[vemel]

I created a new logic for RHS and LHS values, please test my PR. Also, please let me know if output changes work for you, as it changed dramatically in some cases.

Also, give me some ideas when RHS-value should be rendered as a single line and LHS shuld be splitted. THings some to my mind:

• RHS has # type: ignore and LHS does not
• LHS is longer that RHS? - probably, but could lead to unstable output easily.