Re: automake-1.6.1: verbose test failure results: Sun SPARC Solaris 2.8

[Alexandre Duret-Lutz]

>>> "Nelson" == Nelson H F Beebe <address@hidden> writes:

[...]

 >>> Like the cond19.test and subsref.test failures on Solaris 2.7,
 >>> this is a timestamp issue I don't undersand.
 >>> 
 >>> The question is why does the first $MAKE invocation trigger the
 >>> config.status rule.
 >>> 
 >>> How fast is this box?

 Nelson> The box is a four-processor (400MHz UltraSPARC II) Sun Enterprise
 Nelson> E5500 file server with 1GB RAM; its load average is invariably well
 Nelson> below 1.0.

 Nelson> I haven't investigated why the first $MAKE invocation triggers the
 Nelson> config.status rule.

 Nelson> We are running GNU make 3.79.1.

FWIW, I have been able to reproduce this on

  SunOS ghjulia 5.8 Generic_108528-07 sun4u sparc SUNW,Ultra-5_10

from an NFS mounted directory.  The bug does not appear when I
try in /tmp (type 'swap').

Are you using NFS too?


Here is what I did:

  1. unpack today's CVS Automake
  2. run ./configure
  3. print the timestamps of config.status and Makefile 
     (using zsh's stat builtin)

config.status:
  atime   Fri Apr 26 15:53:12
  mtime   Fri Apr 26 15:53:13
  ctime   Fri Apr 26 15:53:13

Makefile:
  atime   Fri Apr 26 15:53:06
  mtime   Fri Apr 26 15:53:07
  ctime   Fri Apr 26 15:53:15

Given that Makefile is created by config.status, Makefile should
be older.  However, only ctimes are in the right order (and
make cares about mtime).

I still have no idea why this occurs.  Also, I've noted the NFS
server did not has it's date synchronized with the client host
on which I was doing this test.  A 6 or 7 seconds gap.  Maybe
that's related.

[...]

-- 
Alexandre Duret-Lutz