Here is a patch for the bug described in
https://sourceforge.net/tracker/?func=detail&atid=105470&aid=1448060&group_id=5470

It follows the latest Martin followup as I understand
it. I'm not sure if multi-lines field handling should
be totally discarded while reading metadata headers (as
implemented), or if it should only be discarded for the
content-type and plural-forms header. If one think this
patch isn't correct, please comment so I can fix it.

Notice that if the second solution was intended, lines
such as 

"#-#-#-#-#  fr.po  #-#-#-#-#\n"

inserted by msgcat/msgmerge will have to be treated
differently (i.e. multi-lines support but skip lines
begining with a #).

Logged In: YES 
user_id=21627

The patch is wrong; it breaks the existing test case
WeirdMetadataTest. I think the "just a single line" rule
should be restricted to Plural-Forms; not sure whether it
should apply to content-type also (but why not).

I'll attach another patch that adds a test case.

Logged In: YES 
user_id=21627

Couldn't get to it before 2.5; forwarding to 2.4.4/2.5.1