MP190_debian_printer.bin:
	name: application/octet-stream
	data: text/plain
	file: application/octet-stream

xdgmime only considers the first 32 bytes of data from the file:
199   for (i = 0; i < 32 && i < len; ++i)•
200     {•
201        if (chardata[i] < 32 && chardata[i] != 9 && chardata[i] != 10 && chardata[i] != 13)•
202          return XDG_MIME_TYPE_UNKNOWN; /* binary data */•
203     }•
204 •
205   return XDG_MIME_TYPE_TEXTPLAIN;•

If they're all printable characters, then it's text/plain.

"file" detects it as a tar archive, but file says:
# pre-POSIX "tar" archives are handled in the C code.

So we'd need to go look into the file code, and see if it can be adapted.

The file code checks for whether something is a tarball by calculating the checksum and checking whether it matches:
https://github.com/file/file/blob/master/src/is_tar.c#L85

We can't do that in shared-mime-info. I've bumped the limit for checking for printable characters in xdgmime:

commit 9c5802b8da56187c5c6abaf70042d14b12d832a9
Author: Bastien Nocera <hadess@hadess.net>
Date:   Tue Sep 5 14:14:53 2017 +0200

    Check further into the file whether it is text or binary
    
    We were only checking 32 bytes into the file, which might not be enough
    for some tar archives with long filenames.
    
    https://bugs.freedesktop.org/show_bug.cgi?id=97372

Anything further would require changes in the implementation of xdgmime that you're using (I'm guessing the one in Qt).

Hi Bastien, are you sure you actually pushed this commit?

$ git show 9c5802b8da56187c5c6abaf70042d14b12d832a9
fatal: bad object 9c5802b8da56187c5c6abaf70042d14b12d832a9

and also cgit doesn't find it: https://cgit.freedesktop.org/xdg/shared-mime-info/commit/?id=9c5802b8da56187c5c6abaf70042d14b12d832a9

@Bastien: ping?

(In reply to Elvis Angelaccio from comment #3)
> Hi Bastien, are you sure you actually pushed this commit?
> 
> $ git show 9c5802b8da56187c5c6abaf70042d14b12d832a9
> fatal: bad object 9c5802b8da56187c5c6abaf70042d14b12d832a9
> 
> and also cgit doesn't find it:
> https://cgit.freedesktop.org/xdg/shared-mime-info/commit/
> ?id=9c5802b8da56187c5c6abaf70042d14b12d832a9

Comment 2 says:
"I've bumped the limit for checking for printable characters in xdgmime"

So the URL would be:
https://cgit.freedesktop.org/xdg/xdgmime/commit/?id=9c5802b8da56187c5c6abaf70042d14b12d832a9

If this is still a problem, then will be in individual shared mime spec implementations. The above fixes the problem for the code used in shared-mime-info's test cases.

@Bastien: thanks for the reply. This is still a problem, so we will probably need to fix it in the Qt implementation.

@David: can you please confirm that in qtbase we would need something similar to Bastien's patch?

(In reply to Elvis Angelaccio from comment #6)
> @Bastien: thanks for the reply. This is still a problem, so we will probably
> need to fix it in the Qt implementation.

FWIW, similar code doesn't exist in glib's GIO implementation, I filed this bug about adding it:
https://bugzilla.gnome.org/show_bug.cgi?id=795544

OK, I just changed the MIME spec to say 128 rather than 32.
(commit ce12f18 in xdg/shared-mime-info)

And I updated Qt accordingly.
https://codereview.qt-project.org/227674

Thanks for the heads up.

@David: thanks for the change, I can confirm that it does fix this issue. :)

(In reply to David Faure from comment #8)
> OK, I just changed the MIME spec to say 128 rather than 32.
> (commit ce12f18 in xdg/shared-mime-info)

Can you at least prefix the commit message in the future? It's absolutely impossible to know what the commit message refers to. It would be good to also add the reasoning behind the change so that the commit log is somewhat self sufficient.

You're right, and I'm usually very careful about that, in fact. Not sure what happened on that day, must have been in a hurry. Sorry about that.