It's relatively easy to identify if a SyntaxError occurs when parsing an fstring expression or not. The idea is to slightly change the error message to start with "f-string: " when it does (in the same way in which SyntaxError messages are produced by the hand-written fstring parser). I have a working PR that produces the following:

$ cat a.py
f'{a $ b}'
$ ./python.exe a.py
  File "/Users/lysnikolaou/Repositories/cpython/a.py", line 1
    (a $ b)
       ^
SyntaxError: f-string: invalid syntax

Thoughts?

I think it's a good idea.

My only reservation would be: if the parsing of f-strings is moved into the parser, would it be possible to maintain the error new messages?

My only reservation would be: if the parsing of f-strings is moved into the parser, would it be possible to maintain the error new messages?

Probably without a single change.

New changeset 2e0a920 by Lysandros Nikolaou in branch 'master':
bpo-41084: Adjust message when an f-string expression causes a SyntaxError (GH-21084)
2e0a920

Should we maybe backport this to 3.9 as well?

I would say yes unless someone sees an obvious impediment

AFAIK, we don't guarantee the stability of error messages, so I think backporting is fine. It's all the better that this is just prefixing something to an existing error message.

New changeset cb0dc52 by Miss Islington (bot) in branch '3.9':
bpo-41084: Adjust message when an f-string expression causes a SyntaxError (GH-21084)
cb0dc52

The change broke the old parser in Python 3.9, I reopen the issue:

AMD64 Arch Linux VintageParser 3.9:
https://buildbot.python.org/all/#/builders/765/builds/129

FAIL: test_format_specifier_expressions (test.test_fstring.TestCase) (str="f'{4:{/5}}'")
FAIL: test_invalid_syntax_error_message (test.test_fstring.TestCase)
FAIL: test_lambda (test.test_fstring.TestCase) (str="f'{lambda x:x}'")
FAIL: test_parens_in_expressions (test.test_fstring.TestCase) (str="f'{,}'")
FAIL: test_parens_in_expressions (test.test_fstring.TestCase) (str="f'{,}'")

New changeset 990efe0 by Lysandros Nikolaou in branch '3.9':
[3.9] bpo-41084: Fix test_fstring failure when using the old parser (GH-21212)
990efe0

Thanks, Victor!

The fix is merged, so I'm closing this again.