Large C++ projects like the llvm compiler consume too many resources when the whole project is parsed by the NetBeans project debugger/database/code model. It can take upwards of 20 minutes to fully parse the project, and it consumes upwards of 600mb of memory. 

I believe the NetBeans should not try to parse the whole project at once, it should be lazy. When the user opens a file, one merely calls the parser to parse the file, then it should parse the headers recursively until everything which needs to be parsed is parsed. This will decrease the time footprint.

The memory footprint can be lowered by releasing already parsed files. So the parser calls files to be parsed, and the parser sees that they are already parsed or parses it. At this point it can place a time stamp on the file. The parser every so often iterates through the files to check the last time a file was used. If the last use of a file is above a certain threshold, the debug-info-database can release all information related with that file. 10 minutes seems like a good number to me.

Finally, the whole application stalls while it is parsing. The CPU meter goes to 100, and nothing can be accomplished. This should not be so. I cannot be certain about the internals, but something like thread priority must exist. The parser thread should be lower than the graphical user interface thread. In fact, everything should be lower than the graphical user interface thread because it's the most important part of a graphical application.

- Thanks,
Jeff Kunkel