Re: svn commit: r26673 - trunk/subversion/libsvn_fs_fs

On Wed, Sep 19, 2007 at 07:18:19AM -0700, Blair Zajac wrote:
> > Followup to r25689, returning (rather than ignoring) an exceptional error.
>
> Hi Malcolm,
>
> I think you mean r25869?
>

Yeah, I fixed that a minute or two later :-)

> > * subversion/libsvn_fs_fs/fs_fs.c
> > (get_and_increment_txn_key_body): If we are unable to update the
> > transaction-current file due to persistent ESTALE errors, return that
> > error to the caller rather than ignoring it.
>
> I copied the body of this function from read_current() in fs_fs.c which
> doesn't have the code you added in this commit. Should it also have it?
>

read_current() returns err directly to the caller, so there's nothing to
leak. Compare with svn_fs_fs__revision_proplist(), which checks err
explicitly before assigning the output parameter.

[Although in that case there's really no good reason not to follow the
read_current() pattern too: I'm pretty sure we don't guarantee the value
of output parameters in the error case.]

Regards,
Malcolm