[Bug preprocessor/78324] Valgrind issues seen with gcc.dg/tree-ssa/builtin-sprintf-2.c

[dmalcolm at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=78324

--- Comment #1 from David Malcolm <dmalcolm at gcc dot gnu.org> ---
Root cause is that the substring loc code isn't set up to cope with
-ftrack-macro-expansion=0, and attempts to handle this location:

../../src/gcc/testsuite/gcc.dg/tree-ssa/builtin-sprintf-2.c:95:1: note: 
 RNG (0,  0,  0, "%hhi", i)
 ^~~

as the location of the literal, rather than:

 RNG (0,  0,  0, "%hhi", i)
                  ^~~~

On re-parsing to locate substring locations, it attempts to parse the 'R' as a
raw string, and so this code within cpp_interpret_string_1 fires:

      if (*p == 'R')
        {
          const uchar *prefix;

          /* Skip over 'R"'.  */
          p += 2;
          prefix = p;
          while (*p != '(')
            p++;

and the issue happens in the "while" loop, as it erroneously walks through this
memory:

(gdb) p strs.m_vec.m_vecdata[0]
$20 = {len = 3, text = 0xc9bcca0 "RNG"}

trying to find the open parenthesis.

It looks like the substring_loc code needs to handle non-default values of
-ftrack-macro-expansion; am continuing to investigate.