On Thu, Nov 01, 2007 at 02:51:55PM +0100, Marco wrote:
On Thu, 01 Nov 2007 09:59:28 +0100, Jens Seidel <jensseidel@users.sf.net> wrote:
I checked it with gcc 3.3, 3.4 4.1, 4.2, 4.3 and it works indeed great and all tests are successful with one exception:
Thanks for your extensive testing ! :-) Indeed I didn't perform any test with gcc 3.3 but from what you pointed out it seems that doesn't support the Mesonides' trick and this cuts away signature deduction of any function object.
Right, maybe you should not even try to support gcc 3.3 if it causes such problems. Maybe you find nevertheless a clean workaround, I don't know. If you need more information such as a callstack just ask ...
Could you suggest me some paper about compiler bugs, unsupported features and possible workarounds.
No sorry. I just dicovered it during my tests. For bugs you could of course check the bug database of GNU. I'm not sure but I think it's useless to report the bug in gcc 3.3 and hope it gets fixed (it's too old!?) but maybe you can still just report it to get a bug number for references.
PS: Please do not create ZIP files from your DOS based system. All files (even source files!) are marked as executable which is obviously wrong.
Jens
Do you mean the zip files I upload into the vault ?
I referred to the ZIP file you referenced in this thread (a tinyurl one). I extracted it via a simple unzip, I cannot image that I did something wrong. Could it be that ZIP archives do not support file permissions at all? I don't know ...
I always create them under linux, so this sounds really odd to me. However I can create .tar.gz archive if this can be helpful. Let me know.
The best solution is probably to get it integrated into sandbox. I have no trouble with wrong file permissions for a testing version but if all archives for the next few years will be broken I suggest to improve the situation. Jens