Incorrect f-string application

[hotpxl]

Python file

a = 1
b = ("{{hello}}" "{}").format(a)

Command

target/debug/ruff check --select UP032 --fix --diff a.py

Vanilla Ruff, no settings, built from master

commit f91bacbb94e1ba0dc56b15afe388fe1f6c92e3e8 (HEAD -> main, origin/main, origin/HEAD)
Author: Charlie Marsh <charlie.r.marsh@gmail.com>
Date:   Fri Aug 25 19:47:15 2023 -0400

The ruff output is

--- a.py
+++ a.py
@@ -1,2 +1,2 @@
 a = 1
-b = ("{{hello}}" "{}").format(a)
+b = ("{{hello}}" f"{a}")

which is incorrect. The old code, when calling .format, collapses the double braces in {{hello}}. But the Ruff version doesn't

[zanieb]

Hi @hotpxl on what version did this used to work? Could you help us determine when this regression was introduced?

[hotpxl]

I don't know if any version used to work, but I can try and git bisect

realized my original description is misleading, corrected the wording

[dhruvmanila]

It seems like the case of implicit string concatenation and as there are no placeholder in the first string, it's not being considered. If you change the code to b = ("{{hello}}{}" "{}").format(a, a), the fix is correct. I guess the fix could be to add the f prefix to all strings within (...).format() call?

[charliermarsh]

Yeah, we probably need to add the f prefix to all segments in the implicit concat, is my guess.

[raylu]

another example:

print('{{ ' '{} {}'.format(1, 2))

$ ruff check --select UP032 test.py --diff
--- test.py
+++ test.py
@@ -1 +1 @@
-print('{{ ' '{} {}'.format(1, 2))
+print('{{ ' f'{1} {2}')

the correct version is print('{ ' f'{1} {2}')

$ ruff --version
ruff 0.0.291

[MichaReiser]

Yeah, we probably need to add the f prefix to all segments in the implicit concat, is my guess.

Won't that potentially trigger another violation because some parts now use f-strings unnecessarily?

[raylu]

surprisingly, no

print(f'{{ ')

$ ruff check . --select F541
test.py:1:7: F541 [*] f-string without any placeholders

but

print(f'{{ ' f'{1} {2}')

$ ruff check . --select F541
[no output]

but even if it did, F541 is autofixable too. just gotta autofix in the right order

[charliermarsh]

No, we don't flag pieces of an implicit concatenation, if at least one segment contains a placeholder. (This is intentional.)

[AlexWaygood]

Looks like this was fixed back in November in #8697 🥳