2014-12-15 17:39 GMT+01:00 Stefan Bodewig <bode...@apache.org>: > On 2014-12-15, Kristian Rosenvold wrote: > >> There is also the issue of determining if the MANIFEST.MF file really >> needs to be the first file in the jar, which puts an interesting >> constraint on the parallelism. I have been unable to understand if the >> jar spec actually requires this or if it's just stuff from the old >> days. > > AFAIR the jar spec does not require it, but many tools do - maybe even > java.util.jar, but I'm not sure. > > When distinguishing a jar from a "normal" zip most tools require the > "jar marker" to be the first zip extra field of the first entry - and > IIRC some tools we've come across required this entry to be the META-INF > directory.
Some of our code in maven goes through some serious hoops to preserve this, so I've been prepared for this although I really don't like it :) In the context of parallel streams it just means there'll be a "root" stream which will be the start of the archive. It might even end up being more efficient than our current algorithm, which scans the input filenames twice to determine if there is manifest-worthy content. My git repo now has a first *crude* working version of the code with a passing testcase, which means I can merge multiple streams. The hacking continues :) K --------------------------------------------------------------------- To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org For additional commands, e-mail: dev-h...@commons.apache.org
