WavPack silently corrupts on Raspberry Pi?, No error on encoding, only on decoding Options

[ktf]

Hi all,

Sometime ago I got myself a Raspberry Pi and I was setting it up to do a lossless codec speed comparison to expand the lossless audio codec comparison I did recently. However, I came across some weird errors and it turned out WavPack was not producing lossless files at all.

To show the results, I've made a short recording and processed it through WavPack, the results are in this zip-file. I took a few settings to see what happened.

wavpack -f: no errors
wavpack -f -x: no errors
wavpack: no errors
wavpack -x: no errors
wavpack -h: missing data or crc errors detected in 12 block(s)!
wavpack -h -x: missing data or crc errors detected in 12 block(s)!
wavpack -hh: missing data or crc errors detected in 12 block(s)!
wavpack -hh -x2: missing data or crc errors detected in 12 block(s)!
wavpack -hh -x4: missing data or crc errors detected in 3 block(s)!
wavpack -hh -x6: missing data or crc errors detected in 3 block(s)!

However, I tested some other songs as well, most produced errors on wavpack -x and wavpack -f -x as well. For example, 30 Seconds to Mars - 100 Suns

wavpack -f: no errors
wavpack -f -x: missing data or crc errors detected in 6 block(s)!
wavpack: no errors
wavpack -x: missing data or crc errors detected in 100 block(s)!
wavpack -h: missing data or crc errors detected in 119 block(s)!
wavpack -h -x: missing data or crc errors detected in 101 block(s)!
wavpack -hh: missing data or crc errors detected in 119 block(s)!
wavpack -hh -x: missing data or crc errors detected in 119 block(s)!
wavpack -hh -x2: missing data or crc errors detected in 119 block(s)!

Anyone who knows what happened here? Faulty built of ARM Debian? Bugs in the WavPack code with respect to the ARM arch?

[saratoga]

Are you running an older version? this bug might apply to you:

http://www.hydrogenaudio.org/forums/index....showtopic=95889

(Although I'm not sure if its present on ARM)

[ktf]

Are you running an older version?

I keep the system updated, probably the built is old? The problem in that thread with hybrid mode though. Furthermore, this problem does not happen on my Ubuntu 12.10 system, which is built from the same repositories.

[bryant]

Thanks for posting these file; there definitely seems to be an issue.

I compared the WavPack files you generated with those generated on a PC and found that the ones with no errors were identical, but the corrupt ones were slightly smaller (like some data was going missing).

When you say this doesn't happen on Ubuntu 12.10 I assume you mean an X86 build and not an ARM build, right? I would be interested in whether building WavPack from SVN head has the same issue (I am not familiar enough with the Raspberry Pi [yet] to know if that's easy).

David

[bryant]

In fact, I'll bet this was fixed with SVN rev 271 which had to do with compilers that default chars to unsigned (which I think ARM does, for some reason).

[ktf]

When you say this doesn't happen on Ubuntu 12.10 I assume you mean an X86 build and not an ARM build, right?

That's the case, I wasn't specific enough. I was comparing Ubuntu on amd64 (regular notebook) which the supported Debian-for-Raspberry distribution, which is ARM. Notebook works, RPBi doesn't.

In fact, I'll bet this was fixed with SVN rev 271 which had to do with compilers that default chars to unsigned (which I think ARM does, for some reason).

I'll check this soon. I haven't installed a toolchain on RBPi yet, that's why I didn't try it right away. Building on the Raspberry isn't hard at all, it just takes a lot of patience as the CPU is very slow.

This post has been edited by ktf: Feb 8 2013, 23:42

[ktf]

I couldn't resist trying immediately Current revision (280) works fine, 270 has the bug and 271 works fine too, so it seems r271 indeed fixed this.

This post has been edited by ktf: Feb 9 2013, 00:07

[bryant]

I couldn't resist trying immediately Current revision (280) works fine, 270 has the bug and 271 works fine too, so it seems r271 indeed fixed this.

Thanks for verifying this!

A WavPack maintenance release is way past due....

[Mr_Rabid_Teddybe...]

A WavPack maintenance release is way past due....

If you can find the time for it I guess things indicate that it might be approaching that point yes....

[bryant]

I am actually working on 4.70 right now.

In addition to the maintenance fixes, I am also including an option to automatically verify during packing (like FLAC's --verify) so that something like the bug shown here will be caught earlier.