want a way to mark files as binary even if they look like text (eg pdf files)

Does it possible to learn bzr to process pdf files as binary?

D:\...1C7.7\Bases\Работа\Бухгалтерия_рабочая>bzr diff -r0..1 ClientBank |more
=== added directory 'ClientBank'
=== added file 'ClientBank/!Руководство пользователя.pdf'
--- ClientBank/!Руководство пользователя.pdf
 1970-01-01 00:00:00 +0000
+++ ClientBank/!Руководство пользователя.pdf
 2009-06-02 10:57:28 +0000
@@ -0,0 +1,3020 @@
%вгПУ-1.2
0000000016 00000 n xref
+0000000811 00000 n
+0000001466 00000 n
+0000001624 00000 n
+0000001802 00000 n
+0000002214 00000 n
+0000002896 00000 n
+0000012793 00000 n
+0000013006 00000 n
+0000013505 00000 n
+0000013720 00000 n
+0000013901 00000 n
+0000033543 00000 n
+0000034338 00000 n
+0000034863 00000 n
+0000047999 00000 n
+0000048205 00000 n
+0000064608 00000 n
+0000065184 00000 n
+0000065538 00000 n
+0000065746 00000 n
+0000000933 00000 n
+0000001444 00000 n
stream446 /Filter /FlateDecode /Length 150 0 R >> 9edcd0ef631130a4fc6e7>]
+H‰b``` ўҐ l ,N ИА
+ eaаpђf ahb pb ЁћФ |шЯC• Љ3 ˜_m TИ3сэ! dбWШ!Бї§еЅЭтЂfЕЉ $S xЉњ ;ЪX…k_q \й“ б )r PVlh ®№фўRС­т#§ЊВ
·-уЄWыд Ћx ‰@ЋinR© Џ–¦ n
ѕуAґ-‹Ѓ§б#Л v €Xџ
 ?Г Ж Ь к)з h8ЬoXМИl$!n!;-№Amюы& ¦ ¦ † ў њ ш j Љ!¶17°3LЉfфLм`Ї#®ЦIЃGЙ]$QЁ дBЕ‰• ЛT›9Љ\%\ @®76
+Ђ

D:\Develop\test>bzr --version
Bazaar (bzr) 1.15
  Python interpreter: C:\Develop\Python25\python.exe 2.5.4
  Python standard library: C:\Develop\Python25\lib
  bzrlib: C:\Develop\Python25\lib\site-packages\bzrlib
  Bazaar configuration: C:\Documents and Settings\Vladimir\Application Data\bazaar\2.0
  Bazaar log file: D:\Docs\.bzr.log

Copyright 2005, 2006, 2007, 2008, 2009 Canonical Ltd.
http://bazaar-vcs.org/

bzr comes with ABSOLUTELY NO WARRANTY. bzr is free software, and
you may use, modify and redistribute it under the terms of the GNU
General Public License version 2 or later.

I wish to add my voice to this.
I suggest a new configuration file like .bzrignore but more like .bzrhandlers which would specify an alternate handler (possibly none!) for filenames matching user defined patterns:

*.pdf /usr/bin/pdfdiff # a magical pdf diffing tool I hope someone will write. :)
*.odf # empty means do not try to diff
*.mov

The best approach IMO would be to support mime-types for individual files, like svn does. That requires file properties though.

It is possible to define end of line conversion in rules file with file name glob basis.

Why not just add another option in the same rules file which specifies if the file should be handled as a binary or not?

Huh, so the gibberish binary bytes that Adobe suggests including in a comment on the second line aren't enough ... interesting!

I wonder what sort of bytes one would need to put there before bzr *would* be convinced? A NUL, maybe?

from bzrlib.textfile:

def text_file(input):
    """Produce a file iterator that is guaranteed to be text, without seeking.
    BinaryFile is raised if the file contains a NUL in the first 1024 bytes.
    """
    first_chunk = input.read(1024)
    if '\x00' in first_chunk:
        raise BinaryFile()
    return IterableFile(chain((first_chunk,), file_iterator(input)))

Precense of anything < 0x20 is probably a better indication. If one want to be extra strict validity could be checked against the current locale. But that is not a solution to the problem stated in the subject.

(excepting NL and CR ofc, and perhaps ESC as well)

Does anyone know how to solve this problem without a bug fix?