untokenize() does not round-trip for code containing line breaks (`\` + `\n`)

[tomasr8]

Bug report

Bug description:

Code which contains line breaks is not round-trip invariant:

import tokenize, io

source_code = r"""
1 + \
    2
"""

tokens = list(tokenize.generate_tokens(io.StringIO(source_code).readline))
x = tokenize.untokenize(tokens)
print(x)
# 1 +\
#     2

Notice that the space between + and \ is now missing. The current tokenizer code simply inserts a backslash when it encounters two subsequent tokens with a differeing row offset:

cpython/Lib/tokenize.py

Lines 179 to 182 in 9c2bb7d

	row_offset = row - self.prev_row
	if row_offset:
	self.tokens.append("\\\n" * row_offset)
	self.prev_col = 0

I think this should be fixed. The docstring of tokenize.untokenize says:

Round-trip invariant for full input:
Untokenized source will match input source exactly

To fix this, it will probably be necessary to inspect the raw line contents and count how much whitespace there is at the end of the line.

CPython versions tested on:

CPython main branch

Operating systems tested on:

Linux

Linked PRs

• gh-125553: Fix backslash continuation in untokenize #126010
• [3.13] gh-125553: Fix backslash continuation in untokenize (GH-126010) #129153
• [3.12] gh-125553: Fix backslash continuation in untokenize (GH-126010) #130579