On Wed, Sep 23, 2009 at 11:22:09AM +0200, Corinna Vinschen wrote: >On Sep 22 21:02, Eric Blake wrote: >>Eric Blake <ebb9 <at> byu.net> writes: >> >>>>>>Cygwin 1.7 is detecting this situation (which is a step up from 1.5 >>>>>>which did the >>rename >>>>>>anyways), but sets errno to EBUSY instead of EINVAL. >>>>> >>>>>Thanks for catching. Feel free to fix the rename function accordingly. >>>> >>>>OK, I'll look into it (I don't know how large the patch will be, yet). >>> >>>And link("a","f/.") should not create "f" as a regular file, either. >>>I'm still looking at where to patch things. >> >>I've got a patch in testing for both of these issues. But while >>looking at path.cc, I've noticed a couple of things: >> >>The code doesn't do a very good job of remembering lengths it has >>already seen. For example, with relative paths, the code in >>normalize_posix_path does cwd.get, then strchr; it seems like since >>cwd.get already knows how many bytes it copied, that a simple API >>modification would pass that information back to the caller so that the >>caller doesn't have to use strchr to find the end of the string. >>Anything we can do to avoid rescanning strings of known length will >>provide speedups in path handling. > >This one might be worth a shot, if it's an easy patch. > >>I'm also wondering whether it is time to finally emulate Linux by >>requiring that when doing pathname resolution of 'a/..', that 'a' >>actually exist (either as a directory or a symlink to directory), >>instead of silently ignoring that part of the string. Should I go >>ahead and spend time working up a patch for this? > >This is something which I have on my TODO list for a long time, but I >never saw a simple way to implement this without losing a lot of the >already lousy performance of the path conversion. Additionally the >path conversion code is already quite complicated and I fear the >unwanted side effects such a change could have. Therefore, I tend to >think of this as a welcome post-1.7.1 change.
Can we discuss these types of things in cygwin-developers, please? It makes searching for decisions/discussions marginally more effective if we discuss design decisions there. I do think, however, that before we start making changes "for efficiency" it might be time to start thinking about doing some performance tests on the DLL. But that's after 1.7.1 is released. cgf -- Problem reports: http://cygwin.com/problems.html FAQ: http://cygwin.com/faq/ Documentation: http://cygwin.com/docs.html Unsubscribe info: http://cygwin.com/ml/#unsubscribe-simple