[Unison-hackers] Unison on a Synology DS412+

[Fabrice Rossi]

Hi,

Le 06/11/2013 17:27, Jérôme Vouillon a écrit :
> On 05/11/2013 15:43, Benjamin C. Pierce wrote:
>>> [...]
>>> So I would recommend to add to the svn repository something similar to
>>> this old glibc workaround.
>>
>> I don't have it on my system either.  If it's working for you, perhaps
>> we should add the version you used to the repo, if it's
>> license-compatible.

If I'm not mistaken, it's a direct copy of an old version of some kernel
header and it has been written John McCutchan (according to the
copyright), the original author of inotify. So it's GPL v2.

> I had planned to remove this compatibility code, as I did not expect it
> to be useful to anyone. This is why inotify_compat.h is not included in
> the svn repository.

I would be glad if you could include (license permitting) the version
I'm proposing as it seems to compile everywhere I've tested it. Synology
tends to be extremely conservative at least for the libc. The last
version of their system (DSM 4.3) released at the end of August this
year is still based on linux 3.2.11 and on glibc 2.3.6 (strange
coupling) on my hardware. Other hardwares include older kernels (2.6.32)
and newer glibc...

Anyway, numerous persons use Unison on those NAS and if you release
unison 2.46.20 with the C based fsmonitor but without the workaround,
they will probably complain :-).

>> [..]
>> On OSX (10.8), the -selftest goes through fine.  So there seems to be
>> a difference in the way the fastercheckUNSAFE option is behaving on
>> linux.  That is, indeed, worth investigating a bit.
>>
>> If you'd like to experiment, have a look at the test that's being run,
>> in the file test.ml.  Even without knowing OCaml, it should be fairly
>> clear what's going on.  If you do the same sequence of file
>> modifications and syncs manually, do you get the same incorrect result?
> 
> What happens is that the file is created and then modified within a
> second. As we use lstat with a one-second granularity, and the file size
> does not change, Unison misses the file update. I don't understand why
> it works on OSX.

I think I understand vaguely the test (thanks to the comments) so if you
need me to test something, I can, but if I understand Jérôme's
explanation, the test should actually not work, should it?

Best regards,

Fabrice