Stefan Fuhrmann wrote on Sun, Oct 21, 2012 at 19:43:16 +0200:
> On Sun, Oct 21, 2012 at 5:55 PM, Daniel Shahaf <d.s_at_daniel.shahaf.name>wrote:
>
> > Stefan Fuhrmann wrote on Sat, Oct 20, 2012 at 15:14:11 +0200:
> > > On Thu, Oct 18, 2012 at 10:22 PM, Daniel Shahaf <d.s_at_daniel.shahaf.name
> > >wrote:
> > >
> > > > Stefan Fuhrmann wrote on Thu, Oct 18, 2012 at 02:56:15 +0200:
> > > > > On Tue, Oct 16, 2012 at 12:55 AM, Daniel Shahaf <
> > d.s_at_daniel.shahaf.name
> > > > >wrote:
> > > > >
> > > > > > stefan2_at_apache.org wrote on Sun, Jun 10, 2012 at 21:34:22 -0000:
> > > > > > > Author: stefan2
> > > > > > > Date: Sun Jun 10 21:34:21 2012
> > > > > > > New Revision: 1348666
> > > > > > >
> > > > > > > URL: http://svn.apache.org/viewvc?rev=1348666&view=rev
> > > > > > > Log:
> > > > > > > When handing out node contents, the delta streams don't need
> > > > > > > to calculate MD5 checksums as the result will not be used and the
> > > > > > > check would be redundant even if it were made.
> > > > > > >
> > > > > > > Thus, rev the svn_txdelta API to calculate the checksum only
> > > > > > > upon specific request and update all callers to use the new API.
> > > > > > >
> > > > > > ...
> > > > > > > Modified: subversion/trunk/subversion/include/svn_delta.h
> > > > > > > URL:
> > > > > >
> > > >
> > http://svn.apache.org/viewvc/subversion/trunk/subversion/include/svn_delta.h?rev=1348666&r1=1348665&r2=1348666&view=diff
> > > > > > >
> > > > > >
> > > >
> > ==============================================================================
> > > > > > > --- subversion/trunk/subversion/include/svn_delta.h (original)
> > > > > > > +++ subversion/trunk/subversion/include/svn_delta.h Sun Jun 10
> > > > 21:34:21
> > > > > > 2012
> > > > > > > @@ -367,10 +367,26 @@ svn_txdelta_md5_digest(svn_txdelta_strea
> > > > > > > + * If @a calculate_checksum
> > > > > > > + * is set, you may call @ref svn_txdelta_md5_digest to get an
> > MD5
> > > > > > checksum
> > > > > > > + * for @a target.
> > > > > > > *
> > > > > > > * Do any necessary allocation in a sub-pool of @a pool.
> > > > > > > + *
> > > > > > > + * @since New in 1.8.
> > > > > > > + */
> > > > > > > +void
> > > > > > > +svn_txdelta2(svn_txdelta_stream_t **stream,
> > > > > > > + svn_stream_t *source,
> > > > > > > + svn_stream_t *target,
> > > > > > > + svn_boolean_t calculate_checksum,
> > > > > >
> > > > > > As of today, only svn_txdelta() and window-test.c pass TRUE for @a
> > > > > > calculate_checksum. Do we anticipate needing to pass TRUE For it
> > in
> > > > any
> > > > > > new code?
> > > > > >
> > > > >
> > > > > No, I don't anticipate extra need for that parameter
> > > > > neither do I feel that it is useless. How would you
> > > > > implement svn_txdelta() without that parameter?
> > > >
> > >
> > > Hm .. wrong question :/
> > >
> > > I would remove the parameter from the public API only. Internally we
> > > > will define both svn_txdelta() and svn_txdelta2() as one-line wrappers
> > > > around some third function which does take an 'svn_boolean_t
> > > > calculate_checksum' parameter.
> > > >
> > >
> > > That would only be o.k., if that internal function was internal
> > > to libsvn_delta. But we would need to call it from other libs
> > > which I see as a strong indicator that others might want this
> > > functionality, too.
> >
> > We wouldn't need to call the internal function from other libs.
> >
> > static void
> > foo_internal(svn_txdelta_stream_t **stream,
> > svn_stream_t *source,
> > svn_stream_t *target,
> > svn_boolean_t calculate_checksum,
> > apr_pool_t *pool);
> >
> > void
> > svn_txdelta2(stream, source, target, pool) {
> > foo_internal(stream, source, target, FALSE, pool);
> > }
> >
> > void
> > svn_txdelta(stream, source, target, pool) {
> > foo_internal(stream, source, target, TRUE, pool);
> > }
> >
>
> We seem to talk past each other here. My problem
> is that the BDB / FSFS libs would need to call the
> *deprecated* svn_txdelta because they need the MD5
> checksum.
Odd. I thought that there were no calls to svn_txdelta() left in the
codebase, and that all calls to svn_txdelta2() passed FALSE for the new
boolean parameter.
Anyway, shrug. I think this thread is too long for the prospective
benefit (one fewer boolean parameter on a 5-ary function).
Cheers
Daniel
Received on 2012-10-22 01:43:47 CEST