clang fails parse stdio.h from (gentoo) glibc-2.6.1

[piotrrak]

Bugzilla Link	3671
Resolution	FIXED
Resolved on	Mar 12, 2010 00:57
Version	unspecified
OS	Linux
Attachments	Reduced testcase.
CC	@DougGregor,@efriedma-quic,@pwo

Extended Description

As Eli Friedman requested in http://lists.cs.uiuc.edu/pipermail/cfe-dev/2009-February/004474.html reporting it here.

Here is a code causing problem:

struct __gconv_step_data {};

typedef struct __gconv_info
{
void * dumb;
extension struct __gconv_step_data __data [];
} *__gconv_t;

typedef union
{
struct __gconv_info __cd;
struct
{
struct __gconv_info __cd;
struct __gconv_step_data __data;
} __combined;
} _G_iconv_t;

reduced.c:15:25: error: variable sized type '__cd' must be at end of struct or class
struct __gconv_info __cd;

union _G_iconv_t definition is in <_G_config.h>, struct _gconv_info in <gconv.h>

[llvmbot]

Reduced further:

struct s {
struct {
int a;
int b[];
} a;
int b;
} s;

[DougGregor]

We need to downgrade (or remove) our diagnostic about the use of structures with flexible array members within other structures. C99 6.7.2.1p16 implies that this code is legal (although accessing s.a.b[0] is undefined behavior).

[llvmbot]

gcc warns with -pedantic

iMac:clang andersca$ gcc -pedantic -fsyntax-only t.c
t.c:4: warning: ISO C90 does not support flexible array members
t.c:5: warning: invalid use of structure with flexible array member

[llvmbot]

I believe C99 6.7.2.1p2 makes it clear that this is illegal; however, language lawyering may be irrelevant on this point.

"A structure or union shall not contain a member with incomplete or function type .., except that the last member of a structure with more than one named member may have incomplete array type; such a structure ... shall not be a member of a structure or an element of an array."

Note that even
struct { int a; struct { int b; int c[]; } b; };
is invalid.

[llvmbot]

*** Bug #4014 has been marked as a duplicate of this bug. ***

[DougGregor]

The Sema part of this is fixed in r66309. Could someone (Daniel? Eli?) check that CodeGen for these cases is okay, then close out this bug?

[efriedma-quic]

CodeGen looks fine; there isn't really anything special to handle.