fix: prevent ValueError crash on blank lines with trailing whitespace (#355)

[apoorvdarshan]

Summary

Fixes #355. Running docformatter on a file that contains a blank line with trailing whitespace crashes in v1.7.8 (a regression introduced by the stdlib tokenize migration in #325):

ValueError: start (4,4) precedes previous end (5,0)

Root cause

A blank line that carries trailing whitespace (e.g. " \n") tokenizes to an NL token whose .line attribute is the whole physical line, " \n" — not the bare "\n"/"\r\n".

is_newline_continuation guarded against blank lines with token.line not in {"\n", "\r\n"}, which only recognizes the exact-blank forms. A whitespace-padded blank line slipped past that guard (and token.line.strip() is "", which is trivially "in" any string), so the blank line was misclassified as a continuation of the preceding line.

That misclassification made _do_update_token_indices rewrite the token's start row back to the previous token's end row, producing a start position that precedes the previous token's end. tokenize.untokenize then rejects the stream with ValueError: start ... precedes previous end ....

Fix

Classify any whitespace-only line as blank by testing the stripped line for content, rather than comparing the raw line against a fixed set of blank strings:

if (
    token.type in (tokenize.NEWLINE, tokenize.NL)
    and token.line.strip()
    and token.line.strip() in prev_token.line.strip()
):
    return True

Genuine continuation lines (whose stripped content is non-empty) are unaffected — this only stops treating blank lines, with or without trailing whitespace, as continuations.

Tests / Verification

• Added regression case issue_355 to tests/formatter/test_do_format_code.py and its data in do_format_code.toml. The source encodes the four trailing spaces on the blank line with explicit \n escapes so they cannot be silently stripped by editors or formatters.
• Confirmed the new test fails on the pristine tree with the exact ValueError from the issue, and passes with the fix applied.
• Full suite green: 540 passed, 3 skipped (the end-to-end subprocess tests require VIRTUAL_ENV to be set in the environment).
• black, isort, and ruff all pass on the changed source.

Disclosure: prepared with AI assistance; reviewed and verified locally.