Date:	Tue, 13 Feb 1996 20:06:57 +0200 (IST)
From:	Eli Zaretskii <eliz AT is DOT elta DOT co DOT il>
To:	"Alexander V. Lukyanov" <lav AT video DOT yars DOT free DOT net>
Cc:	djgpp AT delorie DOT com
Subject:	Re: v2 + rm -r

On Tue, 13 Feb 1996, Alexander V. Lukyanov wrote:

> I've compiled fileutils with djgpp v2 end encountered some trouble
> in recursively deleting directories. rm says that it detected cycle in
> directory tree. I guess it happens because inode equals to hashing and
> the hashing gives the same numbers for several files. Can this be fixed?
> Or maybe I have to modify rm to disable cycle checking?
> 
> I use win95, directories to be deleted are on networked drive.

This has been reported before, but I had no success in getting the
details.  Please help to debug this, because I don't have access to Win95
yet, and on networked drives under DOS this doesn't happen.

The code for `stat' and `fstat' should ensure that no two different files
get the same inode number, even when it falls back to hashing (as indeed
it will under Win95, because Win95 file operations with 32-bit file access
enabled bypass the internal DOS structures which `stat' and `fstat' use to
get at the cluster numbers).  Please try to investigate what exactly are
the requirements that `rm' expects from the inode numbers to comply to,
and how do `stat' or `fstat' fail in this case.  (I would like very much
to solve this without telling people to make DJGPP-specific patches to GNU
packages, if possible.  A lot of effort was invested in making `stat' and
`fstat' compatible to whatever Unix software is used to expect, to avoid
massive patches to file-related programs.)

Thank you in advance.

- Raw text -