flac dev - Dec 2014 - Parts of the standard not supported in reference decoder?

I am working on creating a set of FLAC files for a ?decoder compliance test
corpus? and am looking for examples of things that are part of the standard but
not supported in the reference decoder. I am not looking for things missing from
the reference encoder (such as variable block sizes).

I seem to recall someone on this list (most likely lvqcl or Martijn van Beurden)
mentioning something about a standard-compliant potential improvement to
compression that can?t be pursued at this point because the reference decoder
doesn?t support it.

Any idea what that was? Any other ideas about what might be missing from the
decoder despite being described in the standard?

Maurits

Op 12-12-14 om 14:25 schreef MauritsVB:
> things that are part of the standard but not supported in the reference
decoder. I am not looking for things missing from the reference encoder (such as
variable block sizes).
Isn't this self-contradicting more or less? I mean, FLAC should 
be able to decode every stream it has ever encoded since the 
freeze in FLAC 0.5, so there really should not be anything in 
the standard and in the encoder, but not in the decoder. We 
should have noticed that long ago, anyway.


Op 12-12-14 om 14:25 schreef MauritsVB:
> Any idea what that was? Any other ideas about what might be missing from
the decoder despite being described in the standard?
The only thing that is in the standard but not implemented in 
the reference decoder that I know of is bitdepths other than 8, 
16 or 24 (so, bitdepths like 4 bits, 21 bits and 32 bits for 
example)

I really think that improved compression was me talking about 
variable blocksizes. They are indeed supported by the reference 
decoder, but that functionality has never been implemented in 
the encoder.

On 12 Dec 2014, at 18:05, Martijn van Beurden <mvanb1 at gmail.com> wrote:
> Op 12-12-14 om 14:25 schreef MauritsVB:
>> things that are part of the standard but not supported in the reference
decoder. I am not looking for things missing from the reference encoder (such as
variable block sizes).
> 
> Isn't this self-contradicting more or less? I mean, FLAC should 
> be able to decode every stream it has ever encoded since the 
> freeze in FLAC 0.5, so there really should not be anything in 
> the standard and in the encoder, but not in the decoder. We 
> should have noticed that long ago, anyway.
> 
Of course, I?m not referring to methods supported in the encoder but not the
decoder as that would indeed be noticed quite quickly. I was looking for methods
that might be described in the format/standard but not (yet) implemented in
either encoder or decoder.

My intention is to build a test corpus for decoders which includes various edge
cases. That?s why I was specifically looking for anything missing in the most
common implementations.
> 
> Op 12-12-14 om 14:25 schreef MauritsVB:
>> Any idea what that was? Any other ideas about what might be missing
from the decoder despite being described in the standard?
> 
> The only thing that is in the standard but not implemented in 
> the reference decoder that I know of is bitdepths other than 8, 
> 16 or 24 (so, bitdepths like 4 bits, 21 bits and 32 bits for 
> example)
> 
> I really think that improved compression was me talking about 
> variable blocksizes. They are indeed supported by the reference 
> decoder, but that functionality has never been implemented in 
> the encoder.
Yes, I?m planning to use FLAKE to generate files with variable block sizes to be
added to the test suite.

Anyway, unless anyone has other suggestions of missing implementations, I will
work on the test suite now and post back here once I have things to share.