Robert McGovern wrote:
>>Sorry, scratch that; wrong tests.  Things still not good on Windows.
> 
> 
> Is it a problem with the files not being outputed in binary mode?

That was my first thought, and the to-level API prevents you from 
explicitly opening or writing to zipfile objects in binary mode.  But 
tit appears, from perusing the source, that the under the hood the files 
are opens in binary mode all the time.

I find the some malformed header issue in Linux.  It appears to be an 
issue with how the central and local headers are calculated.

My basic test, on both win32 and Linux, was to create an OpenOffice.doc, 
and extract the inner files (OOo files are just a zip set of some XML 
files).

Rezip the same files using jar, see if OOo can open the results.
Rezip the same files using GNU zip, see if OOo can open the results.
Rezip the same files rubyzip, see if OOo can open the results.

The first two work fine on Linux and win32 (using cygwin's zip; PKZip 
works OK, too.)

The last fails on both Linux and win32, and testing the file with zip -T 
gives the message about the headers.

James