Today I could track down an annoying issue: on some installations of the newer version of RSS Bandit (starting with the first betas of 1.5.0.x) it starts using 100% CPU time on a thread. As we started also to use lucene.NET with this new release I guessed it was a related issue, I was right. But as we run into multiple issues with lucene it was not obvious as a separate problem. So here it is:

We used a slightly modified version of lucene.NET to include the available language dependent analyzers and stopword filters. Before we released, I already fixed some obvious issues I got with e.g. french stemmer. Now we got more with the CJK (Japanese/Korean, bug report) and possibly Chinese analyzers. It looks like it get into a state it never returns. The attached debug callstack was very helpful to got the direction to search for! I found this: TextReader.Read(char[], int, int) was called and the return value checked against -1 only, not against 0 (zero). So I run over all analyzers to find all the places I have to change that code and here it is: the modified lucene.net.dll (zipped, 142K).

To all users that run into the problem: can you test the modification, please? Just download, unzip into the installation folder, restart Bandit. Please report here (or as comments to the bug reports mentioned above) if it help - so we can refresh the installer for the major public.