[Bug c/48418] New: Bit shift operator >>=

[Bug c/48418] New: Bit shift operator >>=

[lisp2d at lisp2d dot net]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48418

           Summary: Bit shift operator >>=
           Product: gcc
           Version: 4.5.1
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: lisp2d@lisp2d.net


int x=1000;
x>>=(sizeof(int)<<3);

x is still 1000

In some cases bit shift operator used with variable (not constant) and compiler
didnot show warning. My opinion is that result must be 0.

Remarked code in my program and code to see what's happened.

/*
ULong::ULong(ULInt    const&x):value2(){
    if(x.uli){
        if(sizeof(unsigned    int)==sizeof(unsigned    long    int)){
            value2.push_back(static_cast<unsigned    int>(x.uli));
            return;}
        UInt    const    ibs(sizeof(unsigned    int)<<3);
        ULInt    x2(x);
        do{
            value2.push_back(static_cast<unsigned    int>(x2.uli));
            (x2.uli)>>=(ibs.ui);
        }while(x2.uli);}}
*/
#include    <iostream>
typedef    struct{unsigned    int    i;}si;
int    main(void){
    unsigned    int    x=1000;
    si    w;
    w.i=sizeof(unsigned    int)<<3;
    std::cout<<"x="<<x<<std::endl;
    x>>=w.i;
    std::cout<<"x>>=w.i -> x="<<x<<std::endl;
    x>>=(sizeof(unsigned    int)<<3);
    std::cout<<"x>>=(sizeof(unsigned    int)<<3) -> x="<<x<<std::endl;
    x>>=32;
    std::cout<<"x>>=32 -> x="<<x<<std::endl;
}

[Bug c/48418] Bit shift operator >>=

[ebotcazou at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48418

Eric Botcazou <ebotcazou at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
                 CC|                            |ebotcazou at gcc dot
                   |                            |gnu.org
         Resolution|                            |WONTFIX

--- Comment #1 from Eric Botcazou <ebotcazou at gcc dot gnu.org> 2011-04-02 21:06:47 UTC ---
> int x=1000;
> x>>=(sizeof(int)<<3);
> 
> x is still 1000

The warning is clear:

t.cc: In function 'void foo()':
t.cc:4:22: warning: right shift count >= width of type

This invokes undefined behavior, any result is acceptable.

> In some cases bit shift operator used with variable (not constant) and compiler
> didnot show warning. My opinion is that result must be 0.

The compiler cannot warn about values computed only at run time.  If the shift
amount is >= width of type at run time, this also invokes undefined behavior.

[Bug c/48418] Bit shift operator >>=

[d.g.gorbachev at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48418

--- Comment #2 from Dmitry Gorbachev <d.g.gorbachev at gmail dot com> 2011-04-02 21:43:15 UTC ---
GCC 4.x regression, for x >= 5:

$ cat > foo.c
unsigned foo(void)
{
#ifdef CONST
  const
#endif
  unsigned i = sizeof(unsigned) << 3;
  unsigned x = 1000;
  return x >> i;
}
^D
$ gcc-4.5.0 -O -S foo.c -DCONST
$ gcc-4.4.6 -O -S foo.c -DCONST
foo.c: In function 'foo':
foo.c:8: warning: right shift count >= width of type

g++ works.

[Bug c/48418] [4.5/4.6/4.7 Regression] Bit shift operator >>=

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48418

Richard Guenther <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|RESOLVED                    |REOPENED
           Keywords|                            |diagnostic
   Last reconfirmed|                            |2011.04.03 11:09:30
                 CC|                            |jsm28 at gcc dot gnu.org
         Resolution|WONTFIX                     |
     Ever Confirmed|0                           |1
            Summary|Bit shift operator >>=      |[4.5/4.6/4.7 Regression]
                   |                            |Bit shift operator >>=
   Target Milestone|---                         |4.5.3

--- Comment #3 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-04-03 11:09:30 UTC ---
I suppose some of the warning code needs to be updated with the C constexpr
changes.  Joseph?

[Bug c/48418] [4.5/4.6/4.7 Regression] Bit shift operator >>=

[joseph at codesourcery dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48418

--- Comment #4 from joseph at codesourcery dot com <joseph at codesourcery dot com> 2011-04-03 11:21:27 UTC ---
It's deliberate that folding of references to const variables is now 
delayed - and ideally it would move out of the front end altogether.  
While decl_constant_value_for_optimization still exists it would be easy 
enough to have a second check for this warning in c_fully_fold_internal, 
but in general the alternatives (which are not mutually exclusive) are 
predictable warnings in the front end that may not catch all cases and may 
sometimes warn in unreachable code, and less-predictable middle-end 
warnings if the problem operation is reachable (here, if a shift by an 
out-of-range constant is found after optimization, which could include 
where the variable with the shift amount isn't marked const).

[Bug c/48418] [4.5/4.6/4.7 Regression] Bit shift operator >>=

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48418

Richard Guenther <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P2

[Bug c/48418] [4.5/4.6/4.7 Regression] Bit shift operator >>=

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48418

Richard Guenther <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|4.5.3                       |4.5.4

--- Comment #5 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-04-28 14:51:30 UTC ---
GCC 4.5.3 is being released, adjusting target milestone.

[Bug c/48418] [4.5/4.6/4.7 Regression] Bit shift operator >>=

[pinskia at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48418

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P2                          |P3
                 CC|                            |rguenth at gcc dot gnu.org

--- Comment #6 from Andrew Pinski <pinskia at gcc dot gnu.org> 2011-08-24 20:47:06 UTC ---
Richard, can you explain why this is a P2? the warning is only a regression in
the sense we would warn about about one case but not any more and the code is
undefined at runtime which means we should warn about it in all cases but
currently don't.

[Bug c/48418] [4.5/4.6/4.7 Regression] Bit shift operator >>=

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48418

Richard Guenther <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P2

--- Comment #7 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-08-25 06:51:10 UTC ---
Because it's a regression and it affects primary languages / targets.  We
shipped with it, so it isn't P1.

[Bug c/48418] [4.5/4.6/4.7/4.8 Regression] Bit shift operator >>=

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48418

Richard Guenther <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|4.5.4                       |4.6.4

--- Comment #8 from Richard Guenther <rguenth at gcc dot gnu.org> 2012-07-02 11:24:40 UTC ---
The 4.5 branch is being closed, adjusting target milestone.

[Bug c/48418] [4.6/4.7/4.8 Regression] Bit shift operator >>=

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48418

--- Comment #9 from Jakub Jelinek <jakub at gcc dot gnu.org> 2013-01-08 14:44:51 UTC ---
Created attachment 29104
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=29104
gcc48-pr48418.patch

Untested fix.  To avoid warning twice, this warns in c_fully_fold_internal
only if orig_op1 isn't INTEGER_CST, but op1 is.  And on the testcase I found
that in 4.8 my SIZEOF_EXPR C++ changes regressed the testcase too.

[Bug c/48418] [4.6/4.7/4.8 Regression] Bit shift operator >>=

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48418

--- Comment #10 from Jakub Jelinek <jakub at gcc dot gnu.org> 2013-01-09 14:51:17 UTC ---
Author: jakub
Date: Wed Jan  9 14:51:09 2013
New Revision: 195051

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=195051
Log:
    PR c/48418
    * c-common.c (c_fully_fold_internal): Warn for LSHIFT_EXPR and
    RSHIFT_EXPR, if orig_op1 isn't INTEGER_CST, op1 is INTEGER_CST
    and is either negative or bigger or equal to type precision
    of the first operand.

    * typeck.c (cp_build_binary_op): For LSHIFT_EXPR and RSHIFT_EXPR,
    call maybe_constant_value for the negative or too big shift
    count warnings.

    * c-c++-common/pr48418.c: New test.

Added:
    trunk/gcc/testsuite/c-c++-common/pr48418.c
Modified:
    trunk/gcc/c-family/ChangeLog
    trunk/gcc/c-family/c-common.c
    trunk/gcc/cp/ChangeLog
    trunk/gcc/cp/typeck.c
    trunk/gcc/testsuite/ChangeLog

[Bug c/48418] [4.6/4.7 Regression] Bit shift operator >>=

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48418

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |jakub at gcc dot gnu.org
            Summary|[4.6/4.7/4.8 Regression]    |[4.6/4.7 Regression] Bit
                   |Bit shift operator >>=      |shift operator >>=

--- Comment #11 from Jakub Jelinek <jakub at gcc dot gnu.org> 2013-01-09 15:03:09 UTC ---
Warning regression fixed for 4.8+ for now.

[Bug c/48418] [4.6/4.7 Regression] Bit shift operator >>=

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48418

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|REOPENED                    |RESOLVED
         Resolution|                            |FIXED

--- Comment #12 from Jakub Jelinek <jakub at gcc dot gnu.org> 2013-02-01 09:11:05 UTC ---
I don't think this is suitable for older branches, for -Werror it might cause
failures where there weren't with older 4.7.x (or 4.6.x) compilers, something
undesirable on release branches if possible.