[C++20] [Modules] Regression module declaration must occur at the start of the translation unit

[ChuanqiXu9]

Reproducer: https://godbolt.org/z/eP84dnKGv

# 1 __FILE__ 1 3
export module a;

I believe this should be accepted no matter what spec says.

[ChuanqiXu9]

CC @yronglin

[llvmbot]

@llvm/issue-subscribers-clang-modules

Author: Chuanqi Xu (ChuanqiXu9)

Reproducer: https://godbolt.org/z/eP84dnKGv

# 1 __FILE__ 1 3
export module a;

I believe this should be accepted no matter what spec says.

[yronglin]

Hmm, according to the wording in P1857R3:

A module directive may only appear as the first preprocessing tokens in a file (excluding the global module fragment.)

and the wording in [cpp.pre]

module-file:
    pp-global-module-fragment[opt] pp-module group[opt] pp-private-module-fragment[opt]

IIUC, # is the first pp-token in the file, so it was rejected by clang. Should line markers be exempted from this rule?

CC @Bigcheese

[Bigcheese]

Hmm, yeah, they really should be exempted from this rule. The goal is to not allow any preprocessor conditionals or most state changes, but these don't fit that.

This line format is an extension so we should just allow them. I'll additionally file and NB comment and write a paper to allow #line.

[yronglin]

Thanks, I'll be happy to follow up on the new rule.

[shafik]

CC @mpark

[AaronBallman]

Hmm, yeah, they really should be exempted from this rule. The goal is to not allow any preprocessor conditionals or most state changes, but these don't fit that.

What kind of state changes are problematic? GNU line markers can change states FWIW (particularly, they change "this is a system header" kind of state).