[vlc-devel] [PATCH 00/10] Item browsing system

[Rémi Denis-Courmont]

Le 2014-05-27 18:39, Jean-Baptiste Kempf a écrit :
>> Currently, SD supports updating nodes asynchronously (that was the 
>> whole
>> point of SD as opposed to playlist demuxers), and a number of SD 
>> plugins do
>> use that feature.
>
> Currently SD is a mess.

That is not true. SD is supposed to deal with dynamic playlists and it 
does that pretty well. I wrote quite a few SD plugins and I have not had 
any problems, so long as browing is not required. There are a few 
historical quirks in the API that nobody bothered to fix. SD was never 
meant to support collapse/expand browsing.

> It does not split the discovery from the browsing,

It was never meant to support browsing. It was only ever meant to 
provide dynamically-generated playlist content.

> and notably cross-protocols, like bonjour/smb, bonjour/ftp, upnp/smb 
> et al.
> Not to mention support for listing of ftp directories.

You don't need SD for that. The tree playlist takes care of 
collapse/expand.

>> You cannot propose your patch series as a solution for the UPnP 
>> problem if
>
> And if SD was supporting updates of nodes asynchronously, why are we 
> at
> the same sorry state of UPnP since 5 years, where everyone who tried
> failed and the only working solution is passing a XSPF?

We are at the same sorry of UPnP because:
1) nobody cared to fix it,
2)there is no interface with *both* browsing *and* dynamic content.

You can do one or the other with the playlist demuxers and SD 
respectively. Adding a new API that also supports only one is a waste of 
time.

>> it removes this basic existing feature. If you want support only for 
>> SMB
>> only on iOS, you can patch the iOS UI and leave the core alone. If 
>> however
>
> SMB is needed for a lot of mobile OSes, and iOS is not the main of 
> it.
> Same for Upnp.

The thing is, there is no need for a new framework to browse file 
systems, *except* for dynamic updates. They are the only missing thing 
at the core level with the current approach. (UI may have issues too, 
but that is a separate problem.)

>> you want a generic solution, you cannot drop the useful features of 
>> the
>> existing solution.
>
> That is sure, but so far, the existing solution is far from good.

The patch set is just a rewrite of the playlist demuxers. 
Functionality-wise, it adds _nothing_.

If you assert that "the existing solution is far from good", then I 
deduce that the patch series is also far from good.

> And the problem with discovery vs browsing is existant.

But that is exactly my point. We need an interface addressing *both* 
aspects simultaneously. In other words, the new interface needs to 
supersede both service discoveries and playlist demuxers. This proposal 
only replaces playlist demuxers.

-- 
Rémi Denis-Courmont