Problem with formatting f-strings on python3.12

[amirstm]

Problem with formatting f-strings on python3.12

[bneradt]

I don't mean to come across as alarmist, but this is a serious bug. The https://github.com/apache/trafficserver project relies upon autopep8 to format its Python files, and this single issue keeping us from moving to fedora:39 because its system version of Python is 3.12. We use a pre-commit hook to block any commits that fail autopep8, so Apache Traffic Server devs currently cannot upgrade to fedora:39. Can the maintainer (@hhatto) please comment on the status of this issue and provide an ETA for the landing of its fix?

Thank you for this tool. It has been helpful to the Apache Traffic Server project.

Kind regards,
Brian

[TheKangaroo]

I see the same problem with the newline after { in f strings but only with the pre-commit v2.0.4 hook.
If run autopep8 v2.0.4 from the CLI everything seems to be okay.
I have the following line in my code:

url=f'https://gitlab.example.example.example.com/api/v4/projects/{project_id}'

With the autopep8 CLI (just the whitespaces will be fixed, e.g. what I would expect):

url = f'https://gitlab.example.example.example.com/api/v4/projects/{project_id}'

With autopep8 pre-commit (the linebreak after { will be inserted):

url = f'https://gitlab.example.example.example.com/api/v4/projects/{
    project_id}'

My pre-commit config is:

  - repo: https://github.com/hhatto/autopep8
    rev: v2.0.4
    hooks:
      - id: autopep8

[bneradt]

@hhatto: Thanks for the reply.

Just in case it's helpful, autopep8 on Python 3.12 greatly changes the white space within format strings, not only the addition of newlines. It adds or subtracts a variety of types of white space in various places. Here's some samples of a run of it on the apache/trafficserver code base:

Removing white space

diff --git a/tests/gold_tests/bad_http_fmt/bad_http_fmt.test.py b/tests/gold_tests/bad_http_fmt/bad_http_fmt.test.py
index 8e84e18e1..059971175 100644
--- a/tests/gold_tests/bad_http_fmt/bad_http_fmt.test.py
+++ b/tests/gold_tests/bad_http_fmt/bad_http_fmt.test.py
@@ -45,7 +45,7 @@ ts.Disk.ip_allow_yaml.AddLines([
     '    action: allow',
     '    methods:',
     '      - GET',
-    f'      - {random_method}',
+    f' - {random_method}',
 ])

Adding spaces around colons:

diff --git a/tests/gold_tests/autest-site/trafficserver.test.ext b/tests/gold_tests/autest-site/trafficserver.test.ext
index 1ebf39e0d..59de90f49 100755
--- a/tests/gold_tests/autest-site/trafficserver.test.ext
+++ b/tests/gold_tests/autest-site/trafficserver.test.ext
@@ -409,9 +409,9 @@ def MakeATSProcess(obj, name, command='traffic_server', select_ports=True,
             port_str += " {ssl_port}:quic {ssl_portv6}:quic:ipv6".format(
                 ssl_port=p.Variables.ssl_port, ssl_portv6=p.Variables.ssl_portv6)
         if enable_proxy_protocol:
-            port_str += f" {p.Variables.proxy_protocol_port}:pp {p.Variables.proxy_protocol_portv6}:pp:ipv6"
+            port_str += f" {p.Variables.proxy_protocol_port}: pp {p.Variables.proxy_protocol_portv6}: pp: ipv6"
             if enable_tls:
-                port_str += f" {p.Variables.proxy_protocol_ssl_port}:pp:ssl {p.Variables.proxy_protocol_ssl_portv6}:pp:ssl:ipv6"
+                port_str += f" {p.Variables.proxy_protocol_ssl_port}: pp: ssl {p.Variables.proxy_protocol_ssl_portv6}: pp: ssl: ipv6"
         #p.Env['PROXY_CONFIG_HTTP_SERVER_PORTS'] = port_str
         p.Disk.records_config.update({
             'proxy.config.http.server_ports': port_str,

Addition of newlines

You mentioned this above:

@@ -73,8 +74,9 @@ tr = Test.AddTestRun()
 tr.Processes.Default.StartBefore(ts)
 tr.Processes.Default.Command = (
     server_cmd(1) +
-    fr" ; printf 'GET {random_path}HTTP/1.0\r\n" +
-    fr"Host: localhost:{Test.Variables.server_port}\r\n" +
+    fr"
+printf 'GET {random_path}HTTP/1.0\r\n" +
+    fr"Host: localhost: {Test.Variables.server_port}\r\n" +
     r"X-Req-Id: 0\r\n\r\n'" +
     f" | nc localhost {ts.Variables.port} >> client.log" +
     " ; echo '======' >> client.log"

Here's a particularly extreme example:

diff --git a/tests/gold_tests/cache/background_fill.test.py b/tests/gold_tests/cache/background_fill.test.py
index 56125a492..c0d02fbb4 100644
--- a/tests/gold_tests/cache/background_fill.test.py
+++ b/tests/gold_tests/cache/background_fill.test.py
@@ -58,7 +58,7 @@ class BackgroundFillTest:
                 "dest_ip=* ssl_cert_name=server.pem ssl_key_name=server.key")
 
             self.ts[name].Disk.records_config.update({
-                "proxy.config.http.server_ports": f"{self.ts[name].Variables.port} {self.ts[name].Variables.ssl_port}:ssl",
+                "proxy.config.http.server_ports": f"{self.ts[name].Variables.port} {self.ts[name].Variables.ssl_port}: ssl",
                 "proxy.config.http.background_fill_active_timeout": "0",
                 "proxy.config.http.background_fill_completed_threshold": "0.0",
                 "proxy.config.http.cache.required_headers": 0,  # Force cache
@@ -111,15 +111,41 @@ class BackgroundFillTest:
         tr = Test.AddTestRun()
         self.__checkProcessBefore(tr)
         tr.Processes.Default.Command = f"""
-curl -X PURGE --http1.1 -vs http://127.0.0.1:{self.ts['for_httpbin'].Variables.port}/drip?duration=4;
-timeout 2 curl --http1.1 -vs http://127.0.0.1:{self.ts['for_httpbin'].Variables.port}/drip?duration=4;
+curl -X PURGE --http1.1 -vs http: //127.0.0.1: {self.ts['for_httpbin'].Variables.port}/drip?duration=4;
+timeout 2 curl --http1.1 -vs http://127.0.0.1:{self.ts['for_httpbin         '                                                        ]
+        .
+                                               V
+                                               a
+                                               r
+                                               i
+                                               a
+                                               b
+                                               l
+                                               e
+                                               s
+                                               .
+                                               p
+                                               o
+                                               r
+                                               t
+                                               } /
+        d
+        r
+        i
+        p
+        ?
+        d
+        u
+        r
+        a
+        t
+        i
+        o
+        n
+        =
+        4
 sleep 4;

Removing newlines

In addition to additional newlines, it seems to also sometimes remove newlines:

-curl --http1.1 -vsk https://127.0.0.1:{self.ts['for_httpbin'].Variables.ssl_port}/drip?duration=4
-"""
-        tr.Processes.Default.ReturnCode = 0
-        tr.Processes.Default.Streams.stderr = "gold/background_fill_1_stderr.gold"
-        self.__checkProcessAfter(tr)
-
+curl --http1.1 -vsk https://127.0.0.1:{self.ts['for_httpbin'].Variables.ssl_port}/drip?d uration=4 """         tr.Processes.Default.ReturnCode = 0         tr.Processes.Default.Streams.stderr = "gold/background_fill_1_stderr.gold"         self.__checkProcessAfter(tr) 

Adding trailing white space

Note that the previous patch actually adds a trailing space at the end, which may not render in github.

[hruskape]

I have been hitting similar problem like @mvo5 . I have had identical version in use autopep8 2.0.4 (pycodestyle: 2.10.0). For me solution was to upgrade pycodestyle to 2.11.1 which is bringing python3.12 support [1].

[1] https://github.com/PyCQA/pycodestyle/blob/main/CHANGES.txt

[jngrad]

The autopep8 v2.1.0 bugfix doesn't seem cover a few edge cases with pycodestyle 2.11.0 and 2.11.1:

• high indentation levels with short strings:

  $ code_sample="if 1:\n    if 1:\n        if 1:\n            err_msg = rf'provided index [{str(n)[1:-1]}] is out of range for shape [{str(list(shape))[1:-1]}]'"
  $ echo -e "${code_sample}"
  if 1:
      if 1:
          if 1:
              err_msg = rf'provided index [{str(n)[1:-1]}] is out of range for shape [{str(list(shape))[1:-1]}]'
  $ echo -e "${code_sample}" | autopep8 -
  if 1:
      if 1:
          if 1:
              err_msg = rf'provided index [{
                  str(n)[1:-1]}] is out of range for shape [{str(list(shape))[1:-1]}]'

• low indentation levels with long strings:

  $ code_sample="if 1:\n    err_msg = f'please provide a test name as argument, like {prefix}lb_cpu__p3m_cpu (got {sys.argv})'"
  $ echo -e "${code_sample}"
  if 1:
      err_msg = f'please provide a test name as argument, like {prefix}lb_cpu__p3m_cpu (got {sys.argv})'
  $ echo -e "${code_sample}" | autopep8 -
  if 1:
      err_msg = f'please provide a test name as argument, like {
          prefix}lb_cpu__p3m_cpu (got {sys.argv})'

• running with the --aggressive mode:

  $ code_sample="class A:\n    def __str__(self):\n        return f'{self.__class__.__name__}({self.selected_mode.__class__.__name__})'"
  $ echo -e "${code_sample}"
  class A:
      def __str__(self):
          return f'{self.__class__.__name__}({self.selected_mode.__class__.__name__})'
  $ echo -e "${code_sample}" | autopep8 -
  class A:
      def __str__(self):
          return f'{self.__class__.__name__}({self.selected_mode.__class__.__name__})'
  $ echo -e "${code_sample}" | autopep8 --aggressive -
  class A:
      def __str__(self):
          return f'{self.__class__.__name__}({
              self.selected_mode.__class__.__name__})'

Environment:

$ python -V
Python 3.12.3
$ autopep8 --version
  autopep8 2.1.0 (pycodestyle: 2.11.1)
$ cat /etc/lsb-release
DISTRIB_ID=Ubuntu
DISTRIB_RELEASE=24.04
DISTRIB_CODENAME=noble
DISTRIB_DESCRIPTION="Ubuntu Noble Numbat (development branch)"

I can drop the --aggressive flag, since there is a proposal to remove it. Yet the other edge cases remain problematic. While Python 3.12 allows line breaks in f-strings (3.12 release notes, section PEP 701: Syntactic formalization of f-strings, paragraph Multi-line expressions and comments), my project supports multiple Python releases, and Python 3.10 and 3.11 raise SyntaxError: unterminated string literal. As a workaround, one can add # nopep8 at the end of impacted f-strings to avoid formatting by autopep8.

[razvan]

It even corrupts sources by adding newlines to fstrings, as can be tested with: https://github.com/stackabletech/image-tools/pull/6/files#diff-cdc0846192c47ce22d8f9a51ee3c0ed9847a222827f12d00a89569b815d92e57R177

[supakeen]

I'm seeing the same behavior where it's corrupting f-strings by adding newlines inside replacements. Using autopep8 2.0.4 and Python 3.12.0.

autopep8: commands[0]> bash -c 'python -m autopep8 --diff --max-line-length 120 -a -a -a -j0 -r --exit-code osbuild/ assemblers/* devices/* inputs/* mounts/* runners/* sources/* stages/*.* stages/test/*.py tools/'
--- original/stages/org.osbuild.debug-shell
+++ fixed/stages/org.osbuild.debug-shell
@@ -31,7 +31,8 @@

     unit = f"""
 [Unit]
-Description=Early root shell on {tty} FOR DEBUGGING ONLY
+Description=Early root shell on {
+        tty} FOR DEBUGGING ONLY
 DefaultDependencies=no
 IgnoreOnIsolate=yes
 ConditionPathExists={tty}

[hhatto]

There are two issues to be discussed and they are listed in order.

1. First Issue

First, @amirstm 's example is incorrect.

$ echo connector = f"socks5://{user}:{password}:{url}:{port}"
connector = fsocks5://{user}:{password}:{url}:{port}
$ echo connector = f"socks5://{user}:{password}:{url}:{port}" | autopep8 -
connector = fsocks5: // {user}: {password}: {url}: {port}

The echo command is not outputting the correct Python f-string example because you did not specify a single quote as an argument.
I recognize the following as correct examples In this case, there is no particular problem with the operation of autopep8.

$ echo 'connector = f"socks5://{user}:{password}:{url}:{port}"'
connector = f"socks5://{user}:{password}:{url}:{port}"
$ echo 'connector = f"socks5://{user}:{password}:{url}:{port}"' | autopep8 -
connector = f"socks5://{user}:{password}:{url}:{port}"

environment:

$ python -V
Python 3.12.0
$ autopep8 --version
autopep8 2.0.4 (pycodestyle: 2.11.1)

2. Second Issue

@razvan @supakeen @bneradt Thanks for reporting.

#712 (comment)
#712 (comment)
#712 (comment)

Second, a line break is inserted at the point after the { in the above f-string where the variable is specified.
This is a surprise to autopep8 users.
We believe this behavior needs to be improved. One idea is to have f-string not change any line breaks, etc.

If you have an opinion on this behavior, please let us know.

Thanks

[supakeen]

@hhatto correct, I personally feel like autopep8 should likely leave any indentation (including newlines) in multi-line strings alone.

There might be some discussion to be had about things that are expected to be formatted inside { and } in f-strings, but adding a newline isn't :)

[miguelmartins3]

Similar issue in this snippet of code:

Hope it helps to troubleshoot or testing.

[zcutlip]

I see the same problem with the newline after { in f strings but only with the pre-commit v2.0.4 hook. If run autopep8 v2.0.4 from the CLI everything seems to be okay. I have the following line in my code:

I'm wondering, is it possible you're running the CLI within a virtualenv with a different version of python?

If I run it from the CLI with my default python, 3.12, it still breaks my f-strings. If I have autopep8 installed in a project's virtualenv with python 3.11 there's no problem

[zcutlip]

Edit: of course this only applies to Mac, and in particular homebrew python, but hopefully it's helpful

If you came here like me in hopes of a workaround, this seems to be working for me:

check if 'brew' exists and if the default python3 version is 3.12, and if so, add python 3.11 to start of $PATH

if _command_exists "brew";
then
    # HACK: autopep8 is currently broken on python 3.12
    # temporarily set python 3.11 at start of PATH
    # https://github.com/hhatto/autopep8/issues/712
    if python3 --version | grep -E '^Python 3\.12' >/dev/null;
    then
        PATH="$(brew --prefix [email protected])"/libexec/bin:$PATH
        export PATH
    fi
fi

[mvo5]

One interesting find is that it seems to be virtualenv inside py312 that is causing the issues. I was trying to write a test case for the (excellent btw) tests that autopep8 has and struggled and then noticed:

$ .tox/autopep8/bin/python --version
Python 3.12.2
$ .tox/autopep8/bin/python -m autopep8 --version
autopep8 2.0.4 (pycodestyle: 2.10.0)
$ .tox/autopep8/bin/python -m autopep8 --diff -r ./test/
/usr/lib/python3.12/site-packages/autopep8.py:182: DeprecationWarning: lib2to3 package is deprecated and may not be able to parse Python 3.10+
  from lib2to3.pgen2 import tokenize as lib2to3_tokenize
...
--- original/./test/mod/test_util_checksum.py
+++ fixed/./test/mod/test_util_checksum.py
@@ -30,5 +30,5 @@
     tempfile.flush()
 
     digest = TEST_RESULT[algorithm]
-    full_digest = f"{algorithm}:{digest}"
+    full_digest = f"{algorithm}: {digest}"
     assert checksum.verify_file(tempfile.name, full_digest), "checksums mismatch"

which looks clearly incorrect. However with the non-virtualenv version:

$ python3 --version
Python 3.12.2
$ python3 -m autopep8 --version
autopep8 2.0.4 (pycodestyle: 2.10.0)
/usr/lib/python3.12/site-packages/autopep8.py:182: DeprecationWarning: lib2to3 package is deprecated and may not be able to parse Python 3.10+
  from lib2to3.pgen2 import tokenize as lib2to3_tokenize
$ 

so here no diff is shown which is expected.

[TheKangaroo]

I'm wondering, is it possible you're running the CLI within a virtualenv with a different version of python?
If I run it from the CLI with my default python, 3.12, it still breaks my f-strings. If I have autopep8 installed in a project's virtualenv with python 3.11 there's no problem

I just checked an my autopep8 installation is via brew, which wraps autopep8 in python 3.11, so no wonder it works as expected:

cat /opt/homebrew/bin/autopep8
───────┬──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────
       │ File: /opt/homebrew/bin/autopep8
───────┼──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────
   1   │ #!/opt/homebrew/opt/[email protected]/bin/python3.11
   2   │ # -*- coding: utf-8 -*-
   3   │ import re
   4   │ import sys
   5   │ from autopep8 import main
   6   │ if __name__ == '__main__':
   7   │     sys.argv[0] = re.sub(r'(-script\.pyw|\.exe)?$', '', sys.argv[0])

For the breaking part, I think pre-commit always runs python in a virtual env, so yes, this matches the observed behaviour of problems with python 3.12 in a virtual env.

[supakeen]

@mvo5 That's a very interesting observation especially as you have the same versions of Python installed in the virtual environment and outside of it. Could you show the output of pipdeptree (you'll have to install it, the package is called pipdeptree as well) in both the virtual environment and outside of it for autopep8?

[mvo5]

AFAICT the output inside the virtualenv and outside are identical (diff shows no difference). It an interesting issue, I really wonder how the fact that it runs inside/outside a virtualenv can have this effect (and why only on py312).

[mvo5]

I have been hitting similar problem like @mvo5 . I have had identical version in use autopep8 2.0.4 (pycodestyle: 2.10.0). For me solution was to upgrade pycodestyle to 2.11.1 which is bringing python3.12 support [1].

[1] https://github.com/PyCQA/pycodestyle/blob/main/CHANGES.txt

Thanks, that is amazing. I think that unblocks us now.

[zcutlip]

will there be a new release of autopep8 in order to trigger an update of pycodstyle?

[hhatto]

autopep8 v2.1.0 has been released.
version 2.1.0 now supports pycodestyle v2.11.0 and above.

[zcutlip]

Can confirm. Several format strings in pyonepassword are getting wrapped per PEP 701. This breaks python 3.9-3.11 compatibility.

I've added # nopep8, but would be nice if it was possible to disable these fixes in a config, or to specify a minimum python version compatibility mode