Re: Translator support for GNU tar

[M. Gerards]

Quoting Marcus Brinkmann <Marcus.Brinkmann@ruhr-uni-bochum.de>:

[...]

> > To add translators to a tar file you have to use the "-a" switch. It
> > doesn't add symlinks, devices and fifos. I guess that is the behaviour
> > we want.
> 
> In effect, yes.  I am not quite sure on tar's internal logic, but it
> appeared to me when I looked at it a long time ago, it was easy to check for
> translators before for everything else (fifos, symlinks etc).

And I did so. The main problem was the stat that fired up the translators. ;)

> Then you can use the Hurd convention of "/hurd/symlink" etc to detect
> symlinks.  This is better because many translators disguise as device files
> or fifos, although they aren't.

Ok, I don't do this ATM, but I can easely ajust my code to do this.

> So, we want to treat real symlinks (/hurd/symlink) as symlinks, etc, but any
> other translator settings as they are.  We don't want to use the standard
> Unix test for this (testing the stat mode bits), but instead use the
> translator string.  This is more accurate.

I promise I will test for symlinks, etc. this way. I'm just curious what can
make things go wrong? Can you give me an example of a case when to store the
node as translator instead of symlink?

> This should not be much harder to implement, and will also make it easier to
> fine-tune the recursive behaviour later.

ok.

I think all my questions are answered right now, with the exception of
file_get_translator and file_set_translator.  Those two functions are
undocumented and I'd like to know if I used them the right way.

Thanks,
Marco