On Fri, 2003-09-05 at 13:32, John Peacock wrote:
> > We cannot set the group permissions of a BDB database without looking
> > under the covers.
> Isn't it possible to use setgid() to change the group ID of the current process
> only (in this case the svnadmin)?
I guess that might work. (For Unix only, of course, and probably only
for repositories created by root.) I'm not sure whether it's worth
adding a flag; anyone who knows to use the flag is probably just as
capable of running chgrp -R after running svnadmin create.
> > libsvn_fs is a library. It cannot in good conscience modify global
> > process state. (We can't even get away with modifying it and setting it
> > back again when we're done, because of threads.)
>
> While it is true that libsvn_fs is a library, it is a mediating library, by
> which I mean that no access to the database happens except through that library,
Correct, but...
> Since all roads lead through
> libsvn_fs, there are no thread issues per se since every thread will set the
> same umask settings before opening the database.
The danger is that the bashed umask might be erroneously applied to
other operations besides Berkeley DB operations.
Also, we wouldn't want to bash the umask in all cases; the user might
actually want a private repository. Determining when it's appropriate
to bash the umask might require violating the BDB abstraction layer.
> Am I misunderstanding umask() (the C library version) then? Does that not
> change the umask for the current process, which would then be inherited when
> calling the BDB library?
You understand correctly, but in the presence of threads, there is no
way to limit the umask change to the BDB operations.
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Received on Fri Sep 5 20:58:18 2003