On Sat, Feb 2, 2013 at 6:00 AM, David Golden <x...@xdg.me> wrote: On Sat, Feb 2, 2013 at 2:20 AM, Gabor Szabo <ga...@szabgab.com> wrote: >> If a real world user can also have similar setup using NFS, then this >> is a legitimate setup, isn't it? >> Isn't this exactly the same situation as Windows vs Linux, where >> Windows might not support all >> the features Linux does? > > +1
As with extending support to other operating systems, sometimes NFS portability is easy and sometimes it's difficult or impossible. If you have to assume that every file system operation might involve an NFS mount, that imposes heavy burdens on certain fields of endeavor. > The more salient question is whether we can identify such things (NFS) > in the report, so, say, something like Andreas' analysis sites could > clearly show the problem is NFS. In a perfect world, CPAN distributions would be able to opt out of supporting NFS as cleanly as they can with operating systems. :P > Separately -- if this is a IPC::Run3/File::Temp vs NFS issue, maybe it > should be fixed to deal more gracefully with an unlinking problem? Dealing with NFS can be a real bear. Cache coherency, absence of delete-on-last-close filehandle semantics, oi... And unlike the OS, you can't pull information out of %Config::Config and change up behavior as necessary. I don't know the specifics of the IPC::Run3/File::Temp issue with NFS, but I have some idea how crazy dealing with race conditions on NFS can be and that's File::Temp's bailiwick. I'll bet the developers have been all over this issue for a long time. Marvin Humphrey
