[libcxx-commits] [libcxx] [libc++] Optimize num_get integral functions (PR #121795)

[Louis Dionne via libcxx-commits]

================
@@ -836,98 +796,137 @@ _InputIterator num_get<_CharT, _InputIterator>::__do_get_signed(
   char_type __atoms1[__atoms_size];
   const char_type* __atoms = this->__do_widen(__iob, __atoms1);
   string __grouping        = this->__stage2_int_prep(__iob, __thousands_sep);
-  string __buf;
-  __buf.resize(__buf.capacity());
-  char* __a     = &__buf[0];
-  char* __a_end = __a;
   unsigned __g[__num_get_base::__num_get_buf_sz];
   unsigned* __g_end = __g;
   unsigned __dc     = 0;
-  for (; __b != __e; ++__b) {
-    if (__a_end == __a + __buf.size()) {
-      size_t __tmp = __buf.size();
-      __buf.resize(2 * __buf.size());
-      __buf.resize(__buf.capacity());
-      __a     = &__buf[0];
-      __a_end = __a + __tmp;
+
----------------
ldionne wrote:

In the new code, we don't seem to have a clear notion of "stage3". If that's the case, you should also change the comment about `Stage 2` above. Please explain what your algorithm does, how it combines `Stage 2` and `Stage 3` together, how and why it differs from what's written in the Standard.

https://github.com/llvm/llvm-project/pull/121795