Kepler

will try this with the JRI implementation of R to see if that resolves the problem. RExpression2 needs more testing as it is!

using RExpression2 (the JRI implementation) works!
of course this requires running from kepler-trunk and launching with java pointing to the appropriate native libraries in the "common" module....

By way of update, Ben partially committed a patch for me that should mostly fix this. If I recall correctly, the remaining code should fix collisions with symbols in record names that are invalid in file names. Specifically, the remaining code addresses the cases where:
1. 2 ports have names that differ by reserved symbols
2. both ports cannot be converted to native tokens and so are saved to disk and passed as file names

The problem arises because the committed code converts almost all non-alphanumeric characters to underscores (_) to create valid file names, creating potential collisions. There is some code to avoid this, but it is implemented before files generated by the firing actor ports are processed. In other words, the existing code can only avoid collisions with ports belonging to different actors and not similarly named ports on a single actor.

I believe the main part of this bug is fixed now. Certainly there are the potential for the temporary filename issues that mare mentioned, but that can be avoided when constructing RExpression actors and their ports. For EML data sources that emit columns with reserved symbols (ie '%CC'), we can now handle that with back ticks (`) and check.names=FALSE.
Can we close this bug [and open a different one to address the port name/temp file issue] so that the bug tracking doesn't drift?

The original summary of the bug has been addressed, so I'm closing this and will open a new one for the additional file collision issue.

Original Bugzilla ID was 3807