rebase keeps last modification time of DLL unchanged

[Christian Franke]

Christopher Faylor wrote:
> On Fri, Mar 09, 2012 at 09:43:07AM +0100, Corinna Vinschen wrote:
>> On Mar  8 21:37, Christian Franke wrote:
>>> Corinna Vinschen wrote:
>>>> On Mar  7 23:07, Christian Franke wrote:
>>>>> The rebase tool does not change last modification timestamp of each
>>>>> DLL even if its data has changed. This is likely because Windows
>>>>> "may" not update the timestamp for files written through a memory
>>>>> mapped view.
>>>>>
>>>>> Is this an intended behavior of rebase?
>>>> Why should rebase change the timestamp?  Apart from the rebasing, the
>>>> DLL is still the same.  If you want to know when it has been last
>>>> rebased, you can look into the file header:
>>>>
>>>>    $ objdump -p cygiconv-2.dll | grep 'Time/Date[^ ]'
>>>>    Time/Date               Tue Mar  6 23:24:12 2012
>>>>
>>> It depends: Changing data without changing st_mtime avoids
>>> (unneeded|required) file copies during incremental backups, rsync,
>>> robocopy, ...
>>>
>>> rebase does not explicitly (re)set the timestamp after rebasing. Is
>>> this by design?
>>>
>>> It relies on weakly defined Windows behavior: "When modifying a file
>>> through a mapped view, the last modification timestamp *may* not be
>>> updated automatically."
>>> http://msdn.microsoft.com/en-us/library/windows/desktop/aa366563.aspx
>> Well, let me put it like this.  Rebase just does its job.  It doesn't
>> actually care for the file timestamp, only for the file header
>> timestamps.  This is not by design, it's just as it is.  So the next
>> question is obvious.  Do you think it should change the timestamp or
>> not?  Why?  A patch is simple and I have it actually already waiting in
>> the scenery.

Both have it its pros and cons, so it depends on user's preferences:
Preserve st_mtime:
+ Incremental Backups are not polluted with unnecessary DLL copies after 
rebaseall is run.

Update st_mtime:
+ Incremental Backups provide an accurate copy (including 
/etc/rebase.db.i386 which matches DLL states)


> I don't think the default should change but maybe an option could be
> added for people who want to see updated times.

Agree.

Actually I had the idea to propose a patch for a new -t(ouch) option. I 
expected some utime()/SetFileTime() call in the source which could be 
simply disabled by the option. After realizing that mtime is preserved 
due to the interesting "timestamp may not be updated" semantics of 
Windows file views, I decided to ask first what the intended behavior of 
rebase is :-)

OT: This also breaks conformance of Cygwin's mmap(). Both POSIX and 
Linux man pages document that st_mtime is updated on writes. There is 
probably no way to fix this.

Christian


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