At 01:10 AM 2/28/2003 -0500, Christopher Faylor wrote: >On Fri, Feb 28, 2003 at 01:02:58AM -0500, Pierre A. Humblet wrote: >>At 12:56 AM 2/28/2003 -0500, you wrote: >>>On Fri, Feb 28, 2003 at 12:49:59AM -0500, Pierre A. Humblet wrote: >>>>OK, following Chris' remarks here is a much smaller set >>>>of changes. >>> >>>Do you think it would make sense to do something along the lines >>>of: >>>>+ path_conv pc (cfd->is_device ? cfd->get_name () : >>cfd->get_win32_name (), PC_SYM_NOFOLLOW); >> >>I guess one could but judging from the times I see in >>strace it's not really justified. >> >>On the other hand that's something that we could look at after you >>integrate your code. There could eventually be a single get_name >>returning what's appropriate. > >This isn't an issue with my code, at least for fstat64. That's one of >the reasons for my changes. I was trying to minimize the number of >duplicate passes through path_conv::check. I still have some tweaking >to do though since adding path_conv to fhandler balloons the sizes of >fhandler_base and ends up using a lot more space in cygheap, which means >more memory to copy on fork, which could mean slower forks, which means >that my performance improvement makes things work more slowly...
Right. The funny part with path_conv is that 2/3 of the storage is in fs_info. Is that really needed/useful? >Anyway, please feel free to check in what you have. OK, this evening. Pierre