Sender:	rich AT phekda DOT freeserve DOT co DOT uk
Message-ID:	<3D139EAB.9DCB25C9@phekda.freeserve.co.uk>
Date:	Fri, 21 Jun 2002 22:46:19 +0100
From:	Richard Dawe <rich AT phekda DOT freeserve DOT co DOT uk>
X-Mailer:	Mozilla 4.77 [en] (X11; U; Linux 2.2.19 i586)
X-Accept-Language:	de,fr
MIME-Version:	1.0
To:	djgpp-workers AT delorie DOT com
Subject:	Re: gcc-3.1: old problem with temporary files
References:	<3D08D753 DOT 9898 DOT 186F2D1 AT localhost> (pavenis AT lanet DOT lv) <3D08F9D5 DOT 4065 DOT 4B0BFF AT localhost>
Reply-To:	djgpp-workers AT delorie DOT com

Hello.

pavenis AT lanet DOT lv wrote:
[snip]
> Anyway it seems that it's not only problem. Anyway it seems to need
> debugging as there are still problems:
> 
> TMPDIR=E:/    works
> 
> TMPDIR=E:\    works
> 
> TMPDIR=E:/TMP/    works
> 
> TMPDIR=E:/TMP\   does not work (access in try() fails and as result TMP is
>         tried). Does not work both with currnt CVS version of DJGPP and
>         v2.03 update. In current CVS version  _chmod(file_name,0)  fails
>       (in src/libc/posix/unistd/access.c)

You can reproduce this using tests/libc/posix/unistd/access.c. Try something
like this from bash:

    cd tests/libc/posix/unistd
    make all
    ./access.exe c:/ c:\\ c:/temp c:\\temp c:/temp/ c:/temp\\

You should get something like this:

bash-2.04$ ./access.exe c:/ c:\\ c:/temp c:\\temp c:/temp/ c:/temp\\
access c:/:     F_OK: YES  R_OK: YES  W_OK: YES  X_OK: YES  D_OK: YES
access c:\:     F_OK: YES  R_OK: YES  W_OK: YES  X_OK: YES  D_OK: YES
access c:/temp: F_OK: YES  R_OK: YES  W_OK: YES  X_OK: YES  D_OK: YES
access c:\temp: F_OK: YES  R_OK: YES  W_OK: YES  X_OK: YES  D_OK: YES
access c:/temp/: F_OK: YES  R_OK: YES  W_OK: YES  X_OK: YES  D_OK: YES
access c:/temp\: F_OK: NO   R_OK: NO   W_OK: NO   X_OK: NO   D_OK: NO

I think there's actually a bug in _put_path. There's code to remove a trailing
slash, but it only does that for forward slashes. IMHO it should do this for
backward slashes too. The diff below contains a patch for this. So far I've
only tested it with a demo program, but it should be tested with more
heavy-duty stuff, given how important _put_path is.

I also noticed what I think is a bug in findfirst. It does not restrict the
attrib argument in any way. We need to restrict it to be <= 255, otherwise we
the bits 8-15 end up being required attributes in the DOS call.

The test program tests/libc/dos/ff.c exposes this bug in findfirst, because it
sets the attrib argument to -1. Nothing is found, since nothing can possibly
match the set of required attributes.

There is a patch below, which restricts the attrib argument appropriately in
findfirst.

Thanks, bye, Rich =]

-- 
Richard Dawe [ http://www.phekda.freeserve.co.uk/richdawe/ ]

Index: src/libc/dos/dir/findfirs.c
===================================================================
RCS file: /cvs/djgpp/djgpp/src/libc/dos/dir/findfirs.c,v
retrieving revision 1.2
diff -p -u -3 -r1.2 findfirs.c
--- src/libc/dos/dir/findfirs.c 31 Aug 1996 21:09:32 -0000      1.2
+++ src/libc/dos/dir/findfirs.c 21 Jun 2002 21:36:39 -0000
@@ -23,6 +23,11 @@ findfirst(const char *pathname, struct f
     return -1;
   }
 
+  /* The low 8 bits are the allowable attributes; the next 8 bits
+     are the required attributes. We have no required attributes,
+     so mask off bits 8-15. */
+  attrib &= 0xff;
+
   pathlen = strlen(pathname) + 1;
 
   _put_path(pathname);
Index: src/libc/dos/io/putpath.c
===================================================================
RCS file: /cvs/djgpp/djgpp/src/libc/dos/io/putpath.c,v
retrieving revision 1.7
diff -p -u -3 -r1.7 putpath.c
--- src/libc/dos/io/putpath.c   10 Jun 2001 11:48:23 -0000      1.7
+++ src/libc/dos/io/putpath.c   21 Jun 2002 21:36:40 -0000
@@ -163,7 +163,7 @@ _put_path2(const char *path, int offset)
   /* remove trailing slash if it doesn't
      represent the root directory */
   if (o-2 >= __tb+offset
-      && _farnspeekb(o-1) == '/'
+      && (_farnspeekb(o-1) == '/' || _farnspeekb(o-1) == '\\')
       && _farnspeekb(o-2) != ':')
     o--;

- Raw text -