Clang parses header names incorrectly as strings

[hubert-reinterpretcast]

Clang does not use the header-name grammar (https://eel.is/c++draft/lex.header#nt:header-name) when parsing the quoted form of header name. Instead, it appears to accept string literals in place of a header name.

Clang fails to accept the following; GCC, EDG, and MSVC all accept (https://godbolt.org/z/ThaPrv9e9):

#if __has_include("\")
#endif

Source (<stdin>)

(as above)

Compiler invocation

clang++ -fsyntax-only -std=c++26 -xc++ -

Actual compiler output

<stdin>:1:19: warning: missing terminating '"' character [-Winvalid-pp-token]
    1 | #if __has_include("\")
      |                   ^
<stdin>:1:19: error: expected "FILENAME" or <FILENAME>
<stdin>:1:19: error: invalid token at start of a preprocessor expression
1 warning and 2 errors generated.

Expected compiler output

(clean compile)

Compiler version info (clang++ -v)

clang version 21.0.0git (https://github.com/llvm/llvm-project.git 41b572b99cd5f222330effa323af9c4da8d73e10)
Target: x86_64-unknown-linux-gnu
Thread model: posix
InstalledDir: /opt/wandbox/clang-head/bin
Found candidate GCC installation: /usr/lib/gcc/x86_64-linux-gnu/13
Found candidate GCC installation: /usr/lib/gcc/x86_64-linux-gnu/14
Selected GCC installation: /usr/lib/gcc/x86_64-linux-gnu/14
Candidate multilib: .;@m64
Selected multilib: .;@m64

[llvmbot]

@llvm/issue-subscribers-clang-frontend

Author: Hubert Tong (hubert-reinterpretcast)

Clang does not use the _header-name_ grammar (https://eel.is/c++draft/lex.header#nt:header-name) when parsing the quoted form of header name. Instead, it appears to accept string literals in place of a header name.

Clang fails to accept the following; GCC, EDG, and MSVC all accept (https://godbolt.org/z/ThaPrv9e9):

#if __has_include("\")
#endif

Source (<stdin>)

(as above)

Compiler invocation

clang++ -fsyntax-only -std=c++26 -xc++ -

Actual compiler output

<stdin>:1:19: warning: missing terminating '"' character [-Winvalid-pp-token]
    1 | #if __has_include("\")
      |                   ^
<stdin>:1:19: error: expected "FILENAME" or <FILENAME>
<stdin>:1:19: error: invalid token at start of a preprocessor expression
1 warning and 2 errors generated.

Expected compiler output

(clean compile)

Compiler version info (clang++ -v)

clang version 21.0.0git (https://github.com/llvm/llvm-project.git 41b572b99cd5f222330effa323af9c4da8d73e10)
Target: x86_64-unknown-linux-gnu
Thread model: posix
InstalledDir: /opt/wandbox/clang-head/bin
Found candidate GCC installation: /usr/lib/gcc/x86_64-linux-gnu/13
Found candidate GCC installation: /usr/lib/gcc/x86_64-linux-gnu/14
Selected GCC installation: /usr/lib/gcc/x86_64-linux-gnu/14
Candidate multilib: .;@<!-- -->m64
Selected multilib: .;@<!-- -->m64

[shafik]

I think lex.header p2 seems to allow what clang is doing but I don't know if it is purposeful or not.

Interesting, we have this FIXME:

llvm-project/clang/lib/Lex/PPDirectives.cpp

Lines 1860 to 1866 in 06e2fd9

	// FIXME: Consider warning on some of the cases described in C11 6.4.7/3 and
	// C++20 [lex.header]/2:
	//
	// If `"`, `'`, `\`, `/*`, or `//` appears in a header-name, then
	// in C: behavior is undefined
	// in C++: program is conditionally-supported with implementation-defined
	// semantics

[hubert-reinterpretcast]

I think lex.header p2 seems to allow what clang is doing but I don't know if it is purposeful or not.

The implementation-defined semantics cannot be taken as causing \ to "escape" the following " because there is no provision for " in a q-char-sequence.