Re: "Perfect" MP3 fails to import

On Thu, 15 Apr 2021 at 18:05, Leland <[email protected]> wrote:

> How about using the error callback only while identifying MP3 files and
> then disable it while doing the actual import?  In this case, the file
> imports just fine.  There might be some audio glitch where the error is,
> but I can’t hear it (I didn’t listen to the whole file).
>

We may need to do something to prevent NaNs. We've had problems with NaN
samples in the past, but I've not seen any since we enabled libmad error
checking. NaN sample problems can be hard to identify, but just a single
NaN can cause devastating corruption further down the line.

Steve



>
>
> *From:* Leland <[email protected]>
> *Sent:* Thursday, April 15, 2021 11:27 AM
> *To:* [email protected]
> *Subject:* Re: [Audacity-devel] "Perfect" MP3 fails to import
>
>
>
> Gave it a try butit still throws the Huffman error.
>
>
>
> *From:* James Crook <[email protected]>
> *Sent:* Thursday, April 15, 2021 10:52 AM
> *To:* audacity-devel <[email protected]>
> *Subject:* Re: [Audacity-devel] "Perfect" MP3 fails to import
>
>
>
> With that mp3 file we reach this point in layer3.c, with cachesz == 20 and
> bits_left == -22
>
>   if (cachesz + bits_left < 0)
>     return MAD_ERROR_BADHUFFDATA;  /* big_values overrun */
>
>
> There's a patch here that addresses length issues and changes that code:
>
> https://fossies.org/linux/avidemux/avidemux_plugins/ADM_audioDecoders/ADM_ad_mad/patches/length-check.patch
>
>    635 +     if (cachesz - fakebits < 1)
>
>   636 +       return MAD_ERROR_BADHUFFDATA;
>
>   637       xrptr[1] = MASK1BIT(bitcache, cachesz--) ?
>
>   638         -requantized : requantized;
>
>   639     }
>
>   640 @@ -1155,9 +1221,6 @@ enum mad_error III_huffdecode(struct mad
>
>   641      }
>
>   642    }
>
>   643
>
>   644 *-  if (cachesz + bits_left < 0)*
>
>   645 *-    return MAD_ERROR_BADHUFFDATA;  /* big_values overrun */*
>
>   646 *-*
>
>   647
>
>
>
> The patch looks pretty extensive.
>
>
>
>     From: Kurt Roeckx <[email protected]>
>     Date: Sun, 28 Jan 2018 19:26:36 +0100
>     Subject: Check the size before reading with mad_bit_read
>
>
>     There are various cases where it attempts to read past the end of the
> buffer
>     using mad_bit_read(). Most functions didn't even know the size of the
> buffer
>     they were reading from.
>
> We are using libmad 0.15.1b.
> Debian have  0.15.2b-9 with these fixes from 2018 in.
>
> This would need careful review.  Probably worth contacting Kurt Roeckx
> too, to confirm we're understanding things correctly, and say thank you.
>
>
>
> --James.
>
>
>
>
>
> On Thu, 15 Apr 2021 at 15:48, Steve Fiddle <[email protected]>
> wrote:
>
>
>
>
>
> On Thu, 15 Apr 2021 at 15:41, James Crook <[email protected]> wrote:
>
> Your first message never made it to this list (I just checked the archive).
> What's the link for the mp3?
>
>
>
> You can get it here:
> https://drive.google.com/file/d/1qCPXtsBfFII2NwsSVMZY9k-Lx6sbhlS8/view?usp=sharing
>
>
>
> Steve
>
>
>
>
>
> On Thu, 15 Apr 2021 at 15:35, Steve Fiddle <[email protected]>
> wrote:
>
> Good detective work James.
>
>
>
> Any ideas about the MP3 file (first post of this thread)? Given your
> explanation of the OGG problem, I assume that it is a different issue.
>
>
>
> I'm thinking that it is probably an Audacity bug because the file appears
> to be a valid MP3 file. On the other hand, if the MP3 is malformed (in some
> way that I've not been able to detect), then it's rather more open to
> opinion whether or not it is an Audacity bug, or a bug in the application
> that created it.
>
>
>
> Steve
>
>
>
> On Thu, 15 Apr 2021 at 13:43, James Crook <[email protected]> wrote:
>
> Logged as: https://bugzilla.audacityteam.org/show_bug.cgi?id=2748
>
>
>
> On Thu, 15 Apr 2021 at 13:27, James Crook <[email protected]> wrote:
>
> It's the same infinite loop as JUCE had:
>
>
> https://forum.juce.com/t/ogg-vorbis-file-puts-juce-oggreader-in-infinite-loop/41148
>
> So will be fixed if we upgrade to 1.3.7 or later.
>
>
>
> On Thu, 15 Apr 2021 at 13:00, James Crook <[email protected]> wrote:
>
> On both 3.0.2 and 2.4.2 it's stuck in library file, vorbisfile.c
>
> static int _fseek64_wrap(FILE *f,ogg_int64_t off,int whence){
>   if(f==NULL)return(-1);
>   return fseek(f,off,whence);
> }
>
> 'off' and 'whence' are both zero ( SEEK_SET ==0, SEEK_CUR==1,
> SEEK_END==2).  We're compiled 32 bit on windows.  Anyone know if it hangs
> on mac/linux (compiled 64 bit) too?
>
>
>
> --James.
>
>
>
> On Thu, 15 Apr 2021 at 12:31, James Crook <[email protected]> wrote:
>
> Confirmed freezes for me on Windows.
>
>
>
> On Thu, 15 Apr 2021 at 12:20, Steve Fiddle <[email protected]>
> wrote:
>
> And by "coincidence" (?), a problem reported importing this ogg file:
>
>
> https://upload.wikimedia.org/wikipedia/commons/0/0f/Claude_Debussy_-_Premi%C3%A8re_Arabesque_-_Patrizia_Prati.ogg
>
>
>
> As with the MP3 file, it can be imported if you select the
> "FFmpeg-compatible files" filter, but in this case, using Audacity's native
> OGG importer causes Audacity to freeze on both Windows and Linux (not
> tested on macOS).
>
>
>
> I don't like coincidences as they may be indicating a deeper problem.
>
>
>
> Steve
>
>
>
> On Thu, 15 Apr 2021 at 12:01, Steve Fiddle <[email protected]>
> wrote:
>
> I've tested the attached file with multiple tools in Linux and it appears
> to be perfectly formed, yet Audacity still complains that it is malformed
> and refuses to import it.
>
>
>
> Steve
>
> _______________________________________________
> audacity-devel mailing list
> [email protected]
> https://lists.sourceforge.net/lists/listinfo/audacity-devel
>
> _______________________________________________
> audacity-devel mailing list
> [email protected]
> https://lists.sourceforge.net/lists/listinfo/audacity-devel
>
> _______________________________________________
> audacity-devel mailing list
> [email protected]
> https://lists.sourceforge.net/lists/listinfo/audacity-devel
>
> _______________________________________________
> audacity-devel mailing list
> [email protected]
> https://lists.sourceforge.net/lists/listinfo/audacity-devel
>
> _______________________________________________
> audacity-devel mailing list
> [email protected]
> https://lists.sourceforge.net/lists/listinfo/audacity-devel
>
> _______________________________________________
> audacity-devel mailing list
> [email protected]
> https://lists.sourceforge.net/lists/listinfo/audacity-devel
>

_______________________________________________
audacity-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/audacity-devel