Code review comment for lp:~jameinel/bzr/2.4-fdatasync-ENOTSUP-1075108

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On 11/6/2012 12:43 PM, Vincent Ladeuil wrote:
> I have a bad feeling about ignoring errors there, the bug you're
> fixing is the first occurrence of fdatasync() failing in a very
> specific context. I'd rather trap only that specific case and keep
> failing for the others rather than silently ignoring them (the
> probability that someone notices the mutter() messages in the log
> is pretty much zero).
>

We can trap whatever we feel like, but note that we've done roughly
the same thing for chmod. If it fails, it just gets in a user's way,
rather than actually doing something useful for them.

I'm willing to be swayed, and I was hesitant as well. But the failure
mode here is that Dimiter can't use bzr *at all* because of ENOTSUP.
If we fail to fdatasync, we still have written the data.

I'd like to turn it around. Can you come up with a specific errno that
clearly indicates we have a problem that we should stop on? (That
wouldn't have triggered during the write, etc).

I like being cautious, but I also like not preventing someone from
actually getting their work done until we manage to add one more error
code into an exception clause.

John
=:->

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.12 (Cygwin)
Comment: Using GnuPG with Mozilla - http://www.enigmail.net/

iEYEARECAAYFAlCY0JMACgkQJdeBCYSNAAPdEgCfQzNNtCTVW4vGNVRNE6Sl4A6t
c/IAmQEfqz9NDi+HSojfoOKV1x1r70ec
=4eMi
-----END PGP SIGNATURE-----