Limewire is really getting fabulous. My only complaint about the most recent betas is that the retry/resume features are really wanting, even worse in some respects than older versions of Limewire. The FastTrack applications really have this one right: once you try to download a file, it *never* quits until you tell it to. If all sources to a file are exhausted, then the client periodically searches on the file's hash value attempting to find new sources to restart the download. Once it does, it picks up where it left off. Quit and restart, and all your partially completed files are still there, automatically searching for new sources.

Compare this with Limewire. First, if all the sources for a file are exhausted ("not available") the partially downloaded file is *lost entirely*. Even if you restart it manually, by finding more sources for the same file/same filename/same hash, it will start the download over form the beginning. This is giving up far too easily, and it is a regression: Limewire used to handle this properly before the swarm stuff went in. (I assume the problem is that the file metadata is thrown away as soon as it's declared not available, but this is too soon since people have the ability to restart downloads manually even if it will not be done automatically.)

And even when the program does try to restart a download automatically (from sources available when "download" was first clicked), the incomplete file cache seems to be keyed by hash-filename rather than just hash. So if the new source uses a different filename than the old source, I think a new file gets created and the transfer starts over. What a waste!

These two problems seem easy to fix, and seriously impact the usability of Limewire for transferring large files (and I should really stop bitching and fix them myself, I know). The extra feature needed to bring Limewire up to the robustness of the FastTrack clients is the ability for the client to automatically search for sources in an attempt to restart aborted downloads. This would best be built on the ability to search by hash rather than filename. Does gnutella have this feature?