[MERGE] don't report malformed bundles as NotABundle

[Aaron Bentley]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

John Arbash Meinel wrote:
>>I want it to either succed or else not raise NotABundle.  I don't want
>>to codify that crlf bundles should raise BadBundle, but I can't assert
>>that they don't fail.
> 
> 
> Well, your assertNotRaises catches all other exceptions.

That is deliberate.  This test explicitly permits it read_bundle to
raise any exception it wishes, except BadBundle.  That is all I care
about.  That is specifically what I want to test.

> I would recommend doing:
> try:
>   do_the_thing
> except NotABundle:
>   self.fail('do_the_thing can't raise NotABundle')

If you do that, you'll get a BadBundle thrown, and the test case will
error out.

> All other actions are regular bugs that need to be fixed.

Any other action will not be a bizarre problem like raising NotABundle.

> Well, then you can change that to "line[1:].rstrip('\r\n')"
> 
> I suppose we could have a problem if the message itself used '\r\n' line
> endings. Though I believe we strip them out when we read the commit
> message. (We read in text mode).

Either the whole bundle text is in DOS line endings, or it's not.  I
propose that we add a strip_cr parameter to BundleReader, and if it's
set, strip the CRs in the generator produced by BundleReader._next().

> Another possible fix is to check the line ending of the header line. And
> then assert that all '#' prefixed lines end with that line ending, and
> strip it.
> 
> 
>>I don't think so.  We may have trailing whitespace in our text, e.g.:
>>
>># message:
>>#   This is a message with a lot of
>>#   whitespace in it.
> 
> 
> Well, I'm not sure what you wrote, because there are no trailing
> whitespaces in the above text.

I put two or three spaces after "of".

> Probably your mailer stripped them.

One way of defeating that particular problem would be to have an
end-of-line marker that would not be stripped.  But it's such a rare
case, it's probably simplest to just base64-encode patches that have
trailing ws, or mixed line-endings.

> The only difference I know of, is that RIO puts the first line of a
> multiline string on the same line. Whereas this one puts the first line
> on the next line. (And uses spaces rather than a tab character).
> Is there some other difference that I don't know about?

RIO uses one line per key/value pair, and uses a blank line as the
stanza separator.  I assume we'd do single stanzas, though.

Aaron
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.1 (GNU/Linux)
Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org

iD8DBQFEmYv40F+nu1YWqI0RAh6rAJ4kwasQ2xjtUmsYqyTDNY6RwVnVugCdF1C5
grcCcKWJxdEdM8YxYA5exGY=
=d/hb
-----END PGP SIGNATURE-----