win32: bzr cat is unusable on Windows

[Alexander Belchenko]

bzr command 'cat' is unusable on Windows because it emit content of file to stabdard stdout that opened on win32 in text mode. So all '\n' converted to '\r\n'. It's make it unusable to obtain past version of file because it mangle line-endings. Especially it affects binary files.

Possible solutions:

1. Reopen sys.stdout in binary mode
2. Provide additional flag for command like '--output=filename' to write file content directly to specified file (as alternative to usage 'bzr cat file1 -rN > file2').

This issue also affects 'diff' and 'bundle-revisions' commands.

[John A Meinel (jameinel)]

I did write this plugin:
http://bzr.arbash-meinel.com/plugins/binary_sys/

Which changes sys.stdout, and sys.stderr so that they are switched to binary mode.

I was planning on making something more explicit (commands that need exact preservation would call a function to enable binary mode).

But I haven't gotten around to it yet.
It could be a simple helper function on Command. And a command like cmd_diff could just call 'self.set_binary_output()' before it continued to do more work.

[Alexander Belchenko (bialix)]

John A Meinel пишет:
> I did write this plugin:
> http://bzr.arbash-meinel.com/plugins/binary_sys/
>
> Which changes sys.stdout, and sys.stderr so that they are switched to
> binary mode.
>
> I was planning on making something more explicit (commands that need
> exact preservation would call a function to enable binary mode).
>
> But I haven't gotten around to it yet.
> It could be a simple helper function on Command. And a command like cmd_diff could just call 'self.set_binary_output()' before it continued to do more work.

I'm thinking about another special attribute as part of Command object
definition. Something similar to encoding_type that you did for encoding
stuff.

Per example, commands like cat, diff, bundle could have attribute

 binary_stdout = True

and when outf object is created based on this flag sys.stdout will be
forced to binary mode.

[John A Meinel (jameinel)]

It is fairly important that we make sys.stdout work in binary mode when necessary. I'm marking this as high, though we can decrease the importance if we decide it isn't critical.

[Henry Ludemann (misc-hl)]

The way I did this to test 'bundle' was to add

fileobj = sys.stdout
# On Windows switch to binary output (we don't want new line
# conversion on the output stream)
if sys.platform == "win32":
    import os, msvcrt
    msvcrt.setmode(sys.stdout.fileno(), os.O_BINARY)

This allowed
   bzr bundle >changes.bundle
...go to another branch...
   bzr merge ..\changes.bundle
which previously failed with
   bzr: ERROR: End of line marker was not \n in bzr revision-bundle

[Alexander Belchenko (bialix)]

It's finally happens!