please test: coreutils-5.90-3

[Eric Blake]

Angelo Graziosi <Angelo.Graziosi <at> roma1.infn.it> writes:

> 1)
> Using coreutils-5.90-3 I have observed  that the command 'cp -p' does not
> preserve the timestamp of a file:
> 
> $ ls -lrt
> -rw-r--r-- 1 Administrator Administrators    418 Aug  7 18:55 t.c
> 
> $ cp -p t.c t.cpp
> $ ls -lrt
> -rw-r--r-- 1 Administrator Administrators    418 Aug  7 18:55 t.c
> -rw-r--r-- 1 Administrator Administrators    418 Oct 18 19:05 t.cpp
> 
> Reinstalling coreutilis-5.3.0-9 the files have the same timestamps.

As far as I can tell, this is a bug in cygwin.

5.3.0 made the following sequence of calls:
dest_desc = open(dest,...);
write(dest_desc,...);
close(dest_desc);
utimes(dest,...);
chmod(dest,...);

5.90 makes the following sequence of calls:
dest_desc = open(dest,...);
write(dest_desc,...);
utimes(dest,...);  // at this point, timestamp is correct
fchmod(dest_desc,...);
close(dest_desc);  // oops, timestamp changed

I can't see anything in SUSv3 that lets close() change the mtime of an open 
file descriptor that has not been modified since the last utimes() to the same 
underlying file.

Meanwhile, coreutils does check for futimes, and would use that if cygwin were 
to provide it.  That might help, since the problem stems from the fact that the 
utimes() to an open file is being lost on close(), but an futimes() would only 
work on an open file.

--
Eric Blake



--
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple
Problem reports:       http://cygwin.com/problems.html
Documentation:         http://cygwin.com/docs.html
FAQ:                   http://cygwin.com/faq/