slacker (s@lack.er) on 1/30/09 wrote:
---------------------------
>Howard Chu (hyc@symas.com) on 1/29/09 wrote:
>---------------------------
>>Quite obviously, if you care about the checksum of an MP3 file with embedded ID3
>>tags, you should write an mp3sum program that ignores ID3 tags when it computes its checksum.
>>
>>Duh...
>
>I'm not sure if this is sarcasm or not, so I'll respond. Your method works, but
>then you have to write this type of parser for all the filetypes that you may encounted
>(MP3, MP4, TIFF, etc...). There's no generic approach.

The mp3 file with id3 tags is a specific format of file, and as such the obvious and expected result of a checksum is that it includes the id3 tags. If you want to store metadata outside the audio data, using a separate file to store that in works wonderfully. What do you get with xattrs? Not something obvious, that's for sure (what to do if you *want* to checksum metadata as well?). Nor does it change the fact that id3 tags remain in the file data stream.

Point is: of course there is no generic approach. If you want to work with application specific file formats, yes, you need application specific knowledge. If you think this changes because "metadata" is stored in xattrs rather than in individual files or within the bytestream, you're deluded. Incredibly trivial and basic examples may actually be possible to be managed from the command line with basic xattr editors (of course it would be also possible and actually much easier if the metadata were stored in different files), but that's because you haven't really thought the idea through.

You want _all_ "metadata" for all "data" to be stored in xattrs? What does this even mean really? Is not the metadata also data in a sense? It may also have metadata associated with it; what to do then? You think everything becomes clear if these get separated out into nice little side-streams? What does a wordprocessor format look like, then? Data is text, and metadata is formatting of that text? Oh but formatting is an integral data of a wordprocessor document too. It also has metadata associated? Document history, that's metadata, a set of changes made to the data and metadata of the document. But it also has metadata (date of change, author, etc). Yep, that's going to be trivial to edit all that using command line tools and without any domain specific knowledge.

Anybody who tells you to "put 'metadata' in extended attributes" is basically an imbecile. Yes it *can* be done in some cases, but the truth is that they probably reached their brain power limit in coming up with "put mp3 song title in xattrs" where it sounds like it works OK (but even then doesn't really actually solve anything).