Autofix SIM102 (NestedIfStatements)

[andersk]

I tried to write an autofixer for SIM102 (NestedIfStatements) using LibCST to preserve comments. But I ran into two issues that potentially point to general problems with the with the SourceCodeLocator API and/or the Fix::replacement API. So I’m opening this draft to discuss what the plan should be.

---

Firstly, if the input if statement is indented, the output if body will be misindented:

 while 1:
     while 2:
-        if 3:
-            if 4:
-                5
+        if 3 and 4:
+    5

In this case, SourceCodeLocator::slice_source_code_range fails to strip the indentation from the second and following lines:

if 3:
            if 4:
                5

and Fix::replacement fails to add that indentation.

---

Secondly, if the input if statement is actually an elif, the fix fails:

while 1:
    while 2:
        if 3:
            4
        elif 5:
            if 6:
                7

In this case, SourceCodeLocator::slice_source_code_range returns this invalid Python that LibCST can’t parse:

elif 5:
            if 6:
                7

[charliermarsh]

The latter issue came up in a slightly different form here: #1694.

[charliermarsh]

@andersk - Ok if I poke around at this branch? Gonna see if I can come up with some helpers for the issues you're describing.

[charliermarsh]

@andersk - Ok, I think this works? We do skip a few cases: (1) if there are any comments between the two if statements, those get destroyed in the fix, so we avoid applying the patch in that case (we do preserve comments in the nested if body, so that's fine); and (2) if the resulting statement is too long, we avoid applying it.

[charliermarsh]

Let me know if you have feedback, and how you feel about merging this change.

[andersk]

This has various failure modes.

while True:
    if True:
        if True:
            """this
is valid"""

            """the indentation on
            this line is significant"""

            "this is" \
"allowed too"

            ("so is"
"this for some reason")

[charliermarsh]

Right, ok, good call. I believe I can fix this by leveraging LibCST's indentation field.

[charliermarsh]

Alright, I think the new version works, but it requires a bunch of trickery.

If the statement is indented, we embed it in a dummy function, like:

def f():
  if ...

Then, at the end, we stripe the function. I think this is the most robust way to preserve indentation. It lets us avoid having to do any indentation or dedentation ourselves (but keeps the source code valid for LibCST).

Separately, we also have to handle elif blocks via special-case.

[andersk]

Yeah, that gets the job done I guess.

I cleaned this up a bit more and split out the helper function changes into a series of logical commits suitable for merging with Rebase and merge.