Re: [groff] Help needed: C++11 language lawyer

[Ralph Corderoy]

Hi Branden,

> I have a patch that quiets the sized-deallocation warning, but the
> most obvious solution (to me) causes a different warning.

That I don't think you give?
Can't the new delete call the existing one to do the work?

Whilst reading
http://git.savannah.gnu.org/cgit/groff.git/tree/src/libs/libgroff/new.cpp
...

    void *operator new(size_t size)
    {
      // Avoid relying on the behaviour of malloc(0).
      if (size == 0)
        size++;

That's a shame as programs like valgrind will now know there's a
writable byte so won't catch code writing to the address from malloc(0)
in error.

      char *p = (char *)malloc(unsigned(size + 8));

This could overflow, allocating, say, four bytes instead of the large
size requested.

      if (p == 0) {
        if (program_name) {
          ewrite(program_name);
          ewrite(": ");
        }
        ewrite("out of memory\n");
        _exit(-1);

This is equivalent to _exit(255) and thus looks equivalent to dying from
signal 127 when accessing the calling sh's `$?' variable.  Also, shells
use high unsigned seven-bit values to indicate problems like `command
not found' so most programs should just exit with 0 or 1, with small
values in particular documented cases.

      }
      ((unsigned *)p)[1] = 0;

Let's hope int is never eight bytes as that would then write beyond
malloc(1)'s area.  :-)  memset(p + 4, 0, 4), or uint32_t if groff uses
those, might be better.

      return p + 8;
    }

-- 
Cheers, Ralph.
https://plus.google.com/+RalphCorderoy