[Bug libc/29399] New: Wrong definition of setjmp

[Bug libc/29399] New: Wrong definition of setjmp

[pavel.morozkin at gmail dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=29399

            Bug ID: 29399
           Summary: Wrong definition of setjmp
           Product: glibc
           Version: unspecified
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: libc
          Assignee: unassigned at sourceware dot org
          Reporter: pavel.morozkin at gmail dot com
                CC: drepper.fsp at gmail dot com
  Target Milestone: ---

The setjmp is defined as:
#define setjmp(env)     _setjmp (env)

This is wrong because it breaks the following code:
#include <setjmp.h>

void f(void)
{
    struct { int x; } _setjmp;
    jmp_buf x;
    setjmp(x);
}

In file included from <source>:1:
<source>: In function 'f':
<source>:7:5: error: called object '_setjmp' is not a function or function
pointer
    7 |     setjmp(x);
      |     ^~~~~~
<source>:5:23: note: declared here
    5 |     struct { int x; } _setjmp;
      |                       ^~~~~~~
Compiler returned: 1

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug libc/29399] Wrong definition of setjmp

[sam at gentoo dot org]

https://sourceware.org/bugzilla/show_bug.cgi?id=29399

Sam James <sam at gentoo dot org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |sam at gentoo dot org

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug libc/29399] Wrong definition of setjmp

[ldv at sourceware dot org]

https://sourceware.org/bugzilla/show_bug.cgi?id=29399

Dmitry V. Levin <ldv at sourceware dot org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         Resolution|---                         |INVALID
                 CC|                            |ldv at sourceware dot org
                URL|                            |https://www.gnu.org/softwar
                   |                            |e/libc/manual/html_node/Res
                   |                            |erved-Names.html
             Status|UNCONFIRMED                 |RESOLVED

--- Comment #1 from Dmitry V. Levin <ldv at sourceware dot org> ---
"reserved names include all external identifiers (global functions and
variables) that begin with an underscore (‘_’) [...]. This is so that the
library and header files can define functions, variables, and macros for
internal purposes without risk of conflict with names in user programs."

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug libc/29399] Wrong definition of setjmp

[pavel.morozkin at gmail dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=29399

--- Comment #2 from Pavel M <pavel.morozkin at gmail dot com> ---
Why it was resolved as INVALID? Per C identifier with block scope that begins
with an underscore, followed by a lowercase letter (e.g. _setjmp) is not
reserved. So the user may use _setjmp for its own purposes. Please reconsider.

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug libc/29399] Wrong definition of setjmp

[schwab@linux-m68k.org]

https://sourceware.org/bugzilla/show_bug.cgi?id=29399

Andreas Schwab <schwab@linux-m68k.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Last reconfirmed|                            |2022-07-24
             Status|RESOLVED                    |REOPENED
     Ever confirmed|0                           |1
         Resolution|INVALID                     |---

--- Comment #3 from Andreas Schwab <schwab@linux-m68k.org> ---
The name is only reserved in file scope.

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug libc/29399] Wrong definition of setjmp

[pavel.morozkin at gmail dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=29399

--- Comment #4 from Pavel M <pavel.morozkin at gmail dot com> ---
Another example:
#define _setjmp 0
#include <setjmp.h>

void f(void)
{
    jmp_buf x;
    setjmp(x);
}

<source>:1:17: error: expected identifier or '(' before numeric constant
    1 | #define _setjmp 0
      |                 ^
<source>: In function 'f':
<source>:1:17: error: called object is not a function or function pointer
    1 | #define _setjmp 0
      |                 ^
/usr/include/setjmp.h:62:25: note: in expansion of macro '_setjmp'
   62 | #define setjmp(env)     _setjmp (env)
      |                         ^~~~~~~
Compiler returned: 1

Per C all identifiers that begin with an underscore, followed by a lowercase
letter AND used as macros are non-reserved.

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug libc/29399] Wrong definition of setjmp

[fweimer at redhat dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=29399

Florian Weimer <fweimer at redhat dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |fweimer at redhat dot com

--- Comment #5 from Florian Weimer <fweimer at redhat dot com> ---
I think C11 is ambiguous whether macros like _setjmp have file scope or not and
are thus reserved or not. It really only makes sense to treat macros as having
file scope for the purpose of identifier reserved-ness, so the example in
comment 4 is invalid (but the one in comment 0 is valid).

Has the scope of macros been clarified in later versions of the C standard?

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug libc/29399] Wrong definition of setjmp

[zackw at panix dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=29399

Zack Weinberg <zackw at panix dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |zackw at panix dot com

--- Comment #6 from Zack Weinberg <zackw at panix dot com> ---
I don't think the standard is ambiguous at all.  Its wording could be improved,
but there's only one interpretation that makes sense.

Since at least C11, section 7.1.3p2 reads in part

> If the program declares or defines an identifier in a context in which it is reserved (other than as allowed by 7.1.4), **or defines a reserved identifier as a macro name**, the behavior is undefined.

The clause set off by double asterisks doesn't make any sense unless "a
reserved identifier" is understood broadly: if the program defines an
identifier that is reserved *in any context* as macro name, the behavior is
undefined.

To put it another way, I believe the intent of this sentence is that any
identifier that is reserved in any phase-7 context is also, unconditionally,
reserved for use as a macro name.

---

The program in comment 4 defines a reserved identifier as a macro name, and is
therefore invalid.  The program in comment 0, on the other hand, doesn't define
any macros; it defines a *local variable*, using a name that is reserved for
use by the implementation *only* at file scope.  This clashes with setjmp.h
assuming that the `_setjmp` in the expansion of the `setjmp` macro will refer
to a global symbol.  I think this *is* a bug in glibc's setjmp.h.  The thing
it's doing with macros should be done with __REDIRECT instead.

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug libc/29399] Wrong definition of setjmp

[pavel.morozkin at gmail dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=29399

--- Comment #7 from Pavel M <pavel.morozkin at gmail dot com> ---
Macros don't have file scope. The scope of macro definitions is defined in C11,
6.10.3.5.

The program in comment 4 defines a non-reserved identifier as a macro name, and
is therefore valid.

P.J. Plauger, Standard C Library Name Space Control, JCLT, 1(4), page 251:
> Remember that the user can write macros that begin with an underscore, followed by a lowercase letter.

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug libc/29399] Wrong definition of setjmp

[fweimer at redhat dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=29399

--- Comment #8 from Florian Weimer <fweimer at redhat dot com> ---
(In reply to Zack Weinberg from comment #6)
> I don't think the standard is ambiguous at all.  Its wording could be
> improved, but there's only one interpretation that makes sense.
> 
> Since at least C11, section 7.1.3p2 reads in part
> 
> > If the program declares or defines an identifier in a context in which it is reserved (other than as allowed by 7.1.4), **or defines a reserved identifier as a macro name**, the behavior is undefined.
> 
> The clause set off by double asterisks doesn't make any sense unless "a
> reserved identifier" is understood broadly: if the program defines an
> identifier that is reserved *in any context* as macro name, the behavior is
> undefined.

On the other hand, 6.2.1p1 refuses to define scope for macro names and macro
parameters, so it's hard to see how the an “identifier with file scope” could
refer to a macro name.

I think 7.1.3 leaves something to be desired anyway because it makes it pretty
clear that this is valid:

#define tm_sec puts ("Hello, world!")
#include <time.h>

int
main (void)
{
  tm_sec;
  return 0;
}

But to implement that, we would have to use non-standard preprocessor features
(#pragma push_macro, #pragma pop_macro).

(Historically, structure fields had file scope (similar to some ML flavors),
but that hasn't been true for a long, long time.)

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug libc/29399] Wrong definition of setjmp

[pavel.morozkin at gmail dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=29399

--- Comment #9 from Pavel M <pavel.morozkin at gmail dot com> ---
The term "file scope" (6.2.1/4) is relevant only at translation phase 7. The
"preprocessing directives are executed, macro invocations are expanded" at
translation phase 4. At translation phase 4 there is no "identifier with file
scope" because there is no "file scope".

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug libc/29399] Wrong definition of setjmp

[schwab@linux-m68k.org]

https://sourceware.org/bugzilla/show_bug.cgi?id=29399

--- Comment #10 from Andreas Schwab <schwab@linux-m68k.org> ---
1.7.3#2 does not mention any context for the identifiers that may not be
defined as macros.

-- 
You are receiving this mail because:
You are on the CC list for the bug.