On Wed, May 20, 2020 at 04:34:53PM +0100, Ramsay Jones wrote:
> On 20/05/2020 01:33, Ramsay Jones wrote:
> > On 19/05/2020 01:57, Luc Van Oostenryck wrote:
> >> + int __safe var3 = obj;
> >> + int *ptr4 = &obj;
> >> + int *ptr4 = ptr; // KO
> >
> > ptr4 declared twice - and sparse didn't complain?
Yes, I was surprised by this too.
> Heh, I had a slightly different example in the test case
> for my '{0}' initializer patch (but involving different
> types as well).
>
> I had a quick look at this and tried to use 'git-bisect' to
> isolate the change which broke this. However, I couldn't find
> a version of sparse that worked correctly! :D (I went all the
> way back to v0.4.2 before giving up - several tagged releases
> didn't even compile without some fix-ups, including v0.4.2).
Yes, it's quite annoying when bisecting, but well ...
> Just FYI, this was my test-case:
>
> $ cat -n test-dup-decl.c
> 1 #ifdef WORKS_OK
> 2 static int sobj;
> 3 static int *sptr4 = &sobj;
> 4 static int *sptr4 = 0;
> 5 #endif
> 6
> 7 static void func(void)
> 8 {
> 9 int obj, *ptr;
> 10 int *ptr4 = &obj;
> 11 int *ptr4 = ptr;
> 12 int a;
> 13 float a;
> 14 }
> $
>
> $ gcc -c test-dup-decl.c
> test-dup-decl.c: In function ‘func’:
> test-dup-decl.c:11:7: error: redefinition of ‘ptr4’
> int *ptr4 = ptr;
> ^~~~
> test-dup-decl.c:10:7: note: previous definition of ‘ptr4’ was here
> int *ptr4 = &obj;
> ^~~~
> test-dup-decl.c:13:8: error: conflicting types for ‘a’
> float a;
> ^
> test-dup-decl.c:12:6: note: previous declaration of ‘a’ was here
> int a;
> ^
> $
>
> $ ./sparse test-dup-decl.c
> $
It seems that sparse detect the redefinition when the symbols are global
but not when they're local.
Thanks for noticing this.
-- Luc
