[Patch] More fixes for Martin's issue

[Paolo Carlini]

Hi,

this is the second, final batch, dealing with the initial parsing
of sign, leading zeros, 0, 0x, 0X in the case of integer types.

Interestingly, notwithstanding the numerous correctness fixes
(accompanied by clean ups, eh!), the current performance for
reading back 1,000,000 integers from 0 to 999,999 with:

  std::ifstream in("input_test");
  for (int j, i = 0; i < 10000000; ++i)
    in >> j;

are as follows (P4-2400, -O2):

3.4
---
5.570u 0.070s 0:05.68 99.2%     0+0k 0+0io 203pf+0w

vs

2.96-99
-------
6.080u 0.060s 0:06.18 99.3%     0+0k 0+0io 135pf+0w

vs

3.3.2
-----
41.790u 0.110s 0:42.22 99.2%    0+0k 0+0io 214pf+0w

Tested x86-linux, would like to commit on monday... one of my
last for this year...

Paolo.

/////////////
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: CL_sebor_2
URL: <http://gcc.gnu.org/pipermail/libstdc++/attachments/20031220/456c56df/attachment.ksh>
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: patch_sebor_2
URL: <http://gcc.gnu.org/pipermail/libstdc++/attachments/20031220/456c56df/attachment-0001.ksh>