tests/hardlink1 fails

[Simon Josefsson]

Assar Westerlund <assar at stacken.kth.se> writes:

> Can you try the first appended patch and see if it solves that
> problem?

Yes, thanks!

> > Has anyone experienced problems running cpp on lots of files in afs?
> 
> Sounds unlikely, the hardlink problem is keeping the link count of
> nodes up-to-date.  It might however be caused by inode-number
> collision.  I can think of two ways of figuring out what is going on:
> It might however be caused by inode-number collision.  I can think
> of two ways of figuring out what is going on:
> 
> 1. instrumenting cpp to see why it doesn't want to read a file.

For some reason cpp on my redhat linux box is CCCP (a cpp that is used
in stage1 when building gcc), and it does inode collition detection.

The real CPP in GCC has had inode stuff #ifdef'd out for ages (2.6?),
and should work, I didn't have the patience to wait for gcc to finish
building so I fixed CCCP instead, patch attached for reference if
anyone has experienced similar problems.

> 2. get the -current code and apply the second appended patch to your
> linux kernel and see if that gets rid of the problem.

I haven't tried it yet.

Why doesn't Transarc AFS exhibit this problem?



--- gcc-2.95/gcc/cccp.c~	Tue Jun  1 19:10:01 1999
+++ gcc-2.95/gcc/cccp.c	Sun Aug  8 16:48:38 1999
@@ -5094,6 +5094,7 @@
 	pfatal_with_name (fname);
       *phead = inc;
 
+#if 0
       /* Look for another file with the same inode and device.  */
       if (lookup_ino_include (inc)
 	  && inc->control_macro
@@ -5101,6 +5102,7 @@
 	close (fd);
 	fd = -2;
       }
+#endif
     }
 
     /* For -M, add this file to the dependencies.  */
</#part>