[Vorbis-dev] Patch to stop vcut from generating broken streams

[Michael Smith]

On Saturday 19 June 2004 12:12, Michael Vrable wrote:
> The attached patch is intended to fix the bug in vcut where the
> granulepos on each page is not always set correctly in the output
> stream.  (See, for example, the discussion on vorbis-dev at xiph.org in
> August 2003, "vcut breaks song index ? XMMS search fails".)  It's an
> issue I came up against while editing the recording of a webcast I
> helped organize; I created a workable fix then, and I've tried to clean
> it up now.
>
> The patch should be against the most recent version of vcut from
> svn.xiph.org (not that it seems to be undergoing much active
> development).

Well, yes. vcut isn't built by default, and I've been suggesting for the past
year or so that nobody should ever use it. It was written as proof-of-concept
code, and it successfully proved the concept. It isn't intended for actual
use. The fact that it went into vorbis-tools was, in retrospect, a mistake.


> I've given it some testing, and haven't noticed any broken streams
> created yet.  But there could still be bugs.  One case that I haven't
> tested yet, and which might be still problematic, is splitting a stream
> which starts at a positive granulepos (such as audio saved from the
> middle of a network stream).
>
> I tried to stick to the original coding style as closely as I could.
> (Including what appeared to be 4-space tabs(!).)

They're the only reasonable way to do tabs! :-) (oh, and you missed "80 column
lines", which isn't stuck to as closely as it should be in the existing
source, but is intended...)


>
> I'm not sure if I should be putting myself in the copyright section or
> not; I'm willing to take that out, assign copyrights to Xiph, etc., if
> needed.
>
> It would be nice to see this patch, or some other bug fix, applied to
> the vcut sources.  If there are issues with the patch that should be
> addressed, please let me know.
>

Whilst I expect (without having actually looked at it yet) that this patch
drastically improves things, I don't really want to do anything to vcut that
might suggest to people that it's actually considered _usable_ (because I
don't, even with these fixes, due to a) and b) below.)

Well... unless you actually make it work properly in all cases - at the very
least this requires heavy testing on streams with a) the start cut off, as
you described, and b) chained streams. Neither of which currently work.

Mike