Undiagnosed syntax error (not checking dead #elif conditions)

[llvmbot]

Bugzilla Link	2291
Version	unspecified
OS	NetBSD
Reporter	LLVM Bugzilla Contributor
CC	@lattner,@efriedma-quic

Extended Description

#elif is not in a skipped group so the following must be diagnosed:

#if 1
#elif
#endif

[llvmbot]

Neither gcc nor Comeau emit a warning on this code.

Can this bug be closed or else reference a specification?

[llvmbot]

See the preprocessor grammar in the standard.

GCC does reject it, if you have a recent enough version :)

[efriedma-quic]

*** #4311 has been marked as a duplicate of this bug. ***

[lattner]

Neil, this example seems to directly contradict C99 6.10.1p5:

"Only the first group whose control condition evaluates to true (nonzero) is
processed."

So the "bodies" of the #if/else are not processed, not really relevant to this case.

"Only the first group whose control condition evaluates to true (nonzero) is
processed."

It isn't clear to me, but this sounds like the condition of the #elif in this case should not be "processed". I'm not sure exactly what "processed" means here though. What is your interpretation?

[llvmbot]

Neil, this example seems to directly contradict C99 6.10.1p5:

"Only the first group whose control condition evaluates to true (nonzero) is
processed."

Right. What are the groups here? Looking at the grammar of if-section:

#if 1 // if-group
#elif // elif-groups[opt]
#endif // endif-line

And for if-group:

if constant-expression new-line group[opt]

So in the example the group[opt] associated with the if-group is empty. That is the group that the if controls; and that is the group that is processed [or skipped] according to the value of the expression.

The elif-groups in the grammar is not a "controlled group" of an enclosing true/false conditional and hence can never be skipped.

The grammar and wording are quite clear here I think, and the #elif is processed as normal, which is a syntax error.

[efriedma-quic]

mentioned in #4311

[bwendling]

Neither recent GCCs or Clangs generate a diagnostic here. Is this still an issue that needs addressing?

[llvmbot]

@llvm/issue-subscribers-c

[AaronBallman]

Per DR412, Clang's behavior of not diagnosing the missing #elif condition is correct.