On Tue, Dec 20, 2011 at 5:38 AM, Ludovic Courtès <[email protected]> wrote:
> Hi Nala! > > Thanks for testing! > > Nala Ginrut <[email protected]> skribis: > > > 1. I think file-system-fold based scandir tried to traverse the whole > > directories include sub-directories. It's rather slow for a deep one if I > > just > > want a files list under 0 level directory tree; > > The code had initially approximately 1 typo per line, and I think I’ve > fixed most of them now. ;-) > > So ‘scandir’ does not enter sub-directories. If it does, that’s another > bug. :-) > > > 2. New scandir will crash while encounters a Chinese file name. This will > > be eliminated by using (setlocale LC_ALL "zh_CN.UTF-8"). > > I think it's the same problem we faced in another thread. There's > > something locale problem in Guile. Of course, we have a temporary > solution > > in recent commit; > > Yes, Guile views file names as strings and decodes them from the current > locale encoding. So if there are file names encoded differently, then > scm_from_locale_string, called by ‘readdir’, throws a decoding-error. > > That’s unfortunate, I’m not sure what to do. I think GLib/GIO issues a > warning in such cases, while still being able to handle the file. We > could imagine ‘readdir’ returning a raw bytevector when decoding fails, > and ‘open-file’ & co. could accept it as input. But that’s really ugly. > > I think Mark had some ideas about it, which would be worth checking. > > > 3. It returns weird result. E.g (scandir "mmr") > > ==> ("." "." "." ".." ".." ".." "aa.c" "exclude" "ml" "myecl") > > One of the typos that got fixed, hopefully. :-) > > Can you check again? > > > Anyway, I think new scandir's cool. Though it's little slow than my C > wrap > > version > > Because it uses ‘file-system-fold’, it does one ‘stat’ call for each > file, which the C version doesn’t do. That should be the only > efficiency difference. > > Maybe we could change our ‘scandir’ to return a list of file name/stat > pairs since we have the info anyway? > > Well~it's a good idea I never thought about. And I've tested new commit, it's wonderful. Maybe you should consider this name/stat, now the new scandir is almost 4 times slower than C wrapper version. Anyway, name/stat list would be a smart and easy solution. ;-)
