rsync - Jun 2008 - rsync-3.0.2 -- two build problems

Hi,

when I tried to build rsync-3.0.2 on i686-linux/gnu, I noticed 2 problems:

===================
(1) With the Linux kernel >=2.6.20, "make check" occasionally
fails, due to
subsecond timestamps sometimes being truncated and sometimes being rounded
upwards (both on i686 and x86_64).

Attached are two files demonstrating the problem: rsync-3.0.2-check-out with
the relevant output from "make check", and
rsync-3.0.2-check-ls-full-time
with the output from
 	ls -l --full-time `find exclude -name file4.junk`
(in the testtmp directory, coreutils-6.9).

This same problem occured already for 3.0.0 and 3.0.1 as well as when
rebuilding 2.6.9 with the newer Linux kernel.

All this somehow defeats the purpose of a test suite, and it would be
extremely nice if such discrepancies could be avoided.

===================
(2) I first run
 	../rsync-3.0.2/configure ; make
then
 	make check
and finally (as root)
 	make install

For rsync-3.0.2, "make check" recompiles everything (not just the
special
tools not needed otherwise), and "make install" again compiles
everything.

The later is really bad, because root may not have write access to the build
directory (e.g., on a root-squashed nfs filesystem).

This problem was not present for rsync-3.0.1 or earlier.

===================
It would be really great if both problem could be fixed for 3.0.3.

please reply directly to me, since I am not subscribed to this list.

Regards
Peter Breitenlohner <peb@mppmu.mpg.de>
-------------- next part --------------
check how the directory listings compare with diff:

--- /.th/thsrc/archive/rsync/rsync-3.0.2-build-linux-gnu/testtmp/exclude/ls-from
2008-06-02 18:48:09.000000000 +0200
+++ /.th/thsrc/archive/rsync/rsync-3.0.2-build-linux-gnu/testtmp/exclude/ls-to
2008-06-02 18:48:09.000000000 +0200
@@ -14,9 +14,9 @@
 -rw-r--r--            9   9000.9000        1 2008-06-02 16:48:07
./bar/down/to/foo/.filt2
 -rw-r--r--            7   9000.9000        1 2008-06-02 16:48:07
./bar/down/to/foo/file1
 -rw-r--r--            5   9000.9000        1 2008-06-02 16:48:07
./bar/down/to/foo/file4
--rw-r--r--           18   9000.9000        1 2008-06-02 16:48:08
./bar/down/to/foo/file4.junk
+-rw-r--r--           18   9000.9000        1 2008-06-02 16:48:07
./bar/down/to/foo/file4.junk
 l---------            0      0.0           1                    
./bar/down/to/foo/sym -> too
--rw-r--r--            8   9000.9000        1 2008-06-02 16:48:08
./bar/down/to/foo/to
+-rw-r--r--            8   9000.9000        1 2008-06-02 16:48:07
./bar/down/to/foo/to
 drwxr-xr-x            0   9000.9000        2 2008-06-02 16:48:07
./bar/down/to/foo/too
 drwxr-xr-x            0   9000.9000        3 2008-06-02 16:48:07 ./foo
 -rw-r--r--           23   9000.9000        1 2008-06-02 16:48:07 ./foo/.filt
-------------
check how the files compare with diff:

-------------
----- exclude log ends
FAIL    exclude
-------------- next part --------------
-rw-r--r-- 1 peb THEORY 18 2008-06-02 18:48:08.000000000 +0200
exclude/chk/bar/down/to/foo/file4.junk
-rw-r--r-- 1 peb THEORY 18 2008-06-02 18:48:07.880843508 +0200
exclude/from/bar/down/to/foo/file4.junk
-rw-r--r-- 1 peb THEORY 18 2008-06-02 18:48:07.000000000 +0200
exclude/to/bar/down/to/foo/file4.junk

On Mon, Jun 02, 2008 at 07:31:02PM +0200, Peter Breitenlohner
wrote:
> (1) With the Linux kernel >=2.6.20, "make check" occasionally
fails, due to
> subsecond timestamps sometimes being truncated and sometimes being rounded
> upwards (both on i686 and x86_64).
Rsync doesn't do any rounding, so I don't know where the inconsistency
would be.  We just set the desination time based on the mtime value, and
the tls program outputs an interpretation of the mtime value.
> For rsync-3.0.2, "make check" recompiles everything (not just the
special
> tools not needed otherwise), and "make install" again compiles
everything.
That was only true when building in a separate dir and only if the
proto.h file was older than the newest .c file, but didn't need to be
updated.  I've fixed this in the latest source.

..wayne..

On Tue, 17 Jun 2008, Wayne Davison wrote:
> On Thu, Jun 05, 2008 at 08:22:09AM +0200, Peter Breitenlohner wrote:
>> However, "cp -p" and "touch -r" round to the
nearest second.
>
> Can you try out the attached patch and let me know if it fixes the
> rounding issues?
I have just rebuilt rsync-3.0.2 with your patch (and have removed
proto.h-tstamp from the source tree in order to prevent continuous
recompilation).

Your patch works nicely: "make check" used to have a failure rate of
about
50%; now I had twelve successful "make check" runs.

Thanks a lot
Peter Breitenlohner <peb@mppmu.mpg.de>