gh-139640: Fix swallowing syntax warnings in different modules

[serhiy-storchaka]

Revert GH-131993.

Fix swallowing some syntax warnings in different modules if they accidentally have the same message and are emitted from the same line.

ast.parse() no longer emits syntax warnings for
return/break/continue in finally (see PEP-765) -- they are only emitted during compilation.

• Issue: Syntax warnings can be swallowed in import #139640

[Eclips4]

The AST changes are LGTM.
Thank you Serhiy.

[ncoghlan]

Would it be possible to avoid the side effect of delaying the PEP 765 syntax warnings to the code generation stage? Emitting those syntax warnings during AST construction is one of the key reasons I was comfortable suggesting turning them off globally in #139658 when static code analysis was in use.

[serhiy-storchaka]

Only if you are fine with double warnings in REPL or other places that use ast.parse() + compile() instead of just compile().

There are three kinds of warnings, by the emitter:

• Emitted in the lexer/tokenizer. They are always emitted when you parse the Python sources.
• Emitted in the AST optimizer. They may be emitted twice, in ast.parse() and compile(). Currently, this is only the PEP 765 warnings.
• Emitted in the code generator. They are only emitted if you use compile().

So, you already see not all warnings if you only use ast.parse(), without code generation. The fact that the PEP 765 warnings are emitted in the AST optimizer instead of the code generator is just an implementation detail. It could perhaps even be more convenient to emit them in the code generator.