------- Comment #23 from amylaar at gcc dot gnu dot org 2007-02-05 18:34 ------- (In reply to comment #20) > I suppose PR29323 was found by inspection of GCC code rather than a real-world > testcase so the option to revert that patch on the 4.1 branch looks appealing. > > (CCed Joern to clarify)
It wasn't as clear-cut as this. I was investigating a bug which later turned out to be a duplicate of PR27781. I was searching for the point where the invalid analysis was made, and TREE_NOTHROW was more easily visible than constness (as in ECF_CONST). So, I was working on a real testcase that was affected manifestly by PR27781, and latently by PR29323, i.e. the function was incorrectly tagged as nothrow, although that incorrect piece of analysis had no effect in the original testcase. (Omitting the function call altogether, as was prompted by the invalid constness analysis, had an oversable effect, though - which was the bug I was really working on). -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=29487