Greg Stein wrote on Thu, Jul 14, 2011 at 00:50:21 -0400:
> On Wed, Jul 13, 2011 at 22:39, <danie...@apache.org> wrote:
> >...
> > +++ subversion/branches/fs-progress/subversion/include/svn_fs.h Thu Jul 14
> > 02:39:52 2011
> > @@ -246,6 +246,24 @@ svn_fs_upgrade(const char *path,
> > apr_pool_t *pool);
> >
> > /**
> > + * Callback function type for progress notification.
> > + *
> > + * @a progress is the number of steps already completed, @a total is
> > + * the total number of steps in this stage, @a stage is the number of
> > + * stages (for extensibility), @a baton is the callback baton.
> > + *
> > + * @note The number of stages may vary depending on the backend, library
> > + * version, and so on. @a total may be a best-effort estimate.
> > + *
> > + * @since New in 1.8.
> > + */
> > +typedef void (*svn_fs_progress_notify_func_t)(apr_off_t progress,
> > + apr_off_t total,
> > + int stage,
> > + void *baton,
> > + apr_pool_t *scratch_pool);
>
> How are PROGRESS and TOTAL logically associated with an apr_off_t?
> That type is for file offsets. Progress information wouldn't seem to
> have any correlation. Maybe just a long? Or an apr_int64_t ?
>
I just copied them from svn_ra_progress_notify_func_t. Will fix.
> >...
> > +++ subversion/branches/fs-progress/subversion/include/svn_repos.h Thu Jul
> > 14 02:39:52 2011
> > @@ -242,7 +242,19 @@ typedef enum svn_repos_notify_action_t
> > svn_repos_notify_recover_start,
> >
> > /** Upgrade has started. */
> > - svn_repos_notify_upgrade_start
> > + svn_repos_notify_upgrade_start,
> > +
> > + /** Verifying global data has commenced
> > + * @since New in 1.8. */
> > + svn_repos_notify_verify_aux_start,
>
> Why it is described as "global data", yet the symbol uses "aux"?
>
Because I haven't decided which way to color the bike shed.
> >...
> > @@ -315,6 +327,12 @@ typedef struct svn_repos_notify_t
> > /** For #svn_repos_notify_load_node_start, the path of the node. */
> > const char *path;
> >
> > + /** For #svn_repos_notify_verify_aux_progress;
> > + see svn_fs_progress_notify_func_t. */
> > + apr_off_t progress_progress;
> > + apr_off_t progress_total;
> > + int progress_stage;
>
> See above re: apr_off_t. And should "stage" be an integer, or is that
> an enumerated constant?
>
Dunno. The idea was to not have to revv the API if we ever decide to
add some other checks besides rep-cache.db.
Perhaps even a C string instead of either an int (counter) or an
enum type (which would be backend and library-version specific).
Ideas welcome.
> >...
> > +++ subversion/branches/fs-progress/subversion/libsvn_repos/dump.c Thu Jul
> > 14 02:39:52 2011
> >...
> > @@ -1284,8 +1306,37 @@ svn_repos_verify_fs2(svn_repos_t *repos,
> >
> > /* Verify global/auxiliary data before verifying revisions. */
> > if (start_rev == 0)
> > - SVN_ERR(svn_fs_verify(svn_fs_path(fs, pool), cancel_func, cancel_baton,
> > - pool));
> > + {
> > + struct progress_to_notify_baton ptnb = {
> > + notify_func, notify_baton, NULL
> > + };
> > +
> > + /* Create a notify object that we can reuse within the callback. */
> > + if (notify_func)
> > + ptnb.notify =
> > svn_repos_notify_create(svn_repos_notify_verify_aux_progress,
> > + iterpool);
> > +
> > + /* We're starting. */
> > + if (notify_func)
> > + notify_func(notify_baton,
> > +
> > svn_repos_notify_create(svn_repos_notify_verify_aux_start,
> > + iterpool),
> > + iterpool);
> > +
> > + /* Do the work. */
> > + SVN_ERR(svn_fs_verify(svn_fs_path(fs, iterpool),
> > + (notify_func ? progress_to_notify : NULL),
> > &ptnb,
> > + cancel_func, cancel_baton,
> > + iterpool));
> > +
> > + /* We're finished. */
> > + if (notify_func)
> > + notify_func(notify_baton,
> > +
> > svn_repos_notify_create(svn_repos_notify_verify_aux_end,
> > + iterpool),
> > + iterpool);
> > +
> > + }
>
> It seems the entire block above can be written more clearly with an
> outer-block test of (notify_func), and then a direct call to
> svn_fs_verify() without notify information, or a big block to set up
> and do all the notification stuff. That should be clearer than four
> tests of (notify_func).
>
> Not to mention the conditional setting of .notify, yet there is an
> unconditional usage in progress_to_notify() ... kinda throws you for a
> bit. Until you realize that progress_to_notify() is *also*
> conditionally used.
>
Yeah, I wasn't happy with it; the notifications take up more visual
space than the code. I'll rework it later for simplicity, perhaps along
the lines you mention.
> >...
>
> Cheers,
> -g
Thanks for the review,
