Relatively long-running tests for handling large amounts of data, numerous iterations, etc. May have some NUnit, but probably less pervasive than the smaller unit tests.
Couple that with getting runtime performance figures. Think knitr and R Markdown, perhaps.
* Impact of using the lock mechanism on preserve/unpreserve. Guidelines
* Character vectors: measure, and advise best practices.
* Figures on Linux and Mono
Comments: I've been running a number of utilization cases (plots, regressions, summary) in an automated fashion against a few of the package datasets (Cars93 from MASS, nihills from DAAG, etc..). I'm looking to consistently reproduce the errors I'm getting while WebAPI/WCF hosting the R engine. 1) Sometimes 'geterrormsg' fails and the engine is reported as being in an inconsistent state. I can't find a single trigger for it as it happens at odd and random times. For example, when a parse exception is thrown for a syntax error, but the same error will not fail other times. 2) Stack limit messages. Again this seems random. Admittedly, I'm not happy with the way the C# SEXP object graph inherits from SafeHandle, but ANTS profiling has been inconclusive as to showing it to be an issue. FWIW, I think the graph should be aggregating a specific SafeHandle type, such as SafeHandleZeroOrMinusOneIsInvalid. One of the key concepts for finalizable objects is that you try to limit classes that end up being stuck in the finalizer queue to the smallest type you can. Aggregating the type addresses this concern as well as other concerns I have about object responsability (mentioned in a different comment). I'm not sure what the concern is about character vectors. Can that be elaborated a bit?
Couple that with getting runtime performance figures. Think knitr and R Markdown, perhaps.
* Impact of using the lock mechanism on preserve/unpreserve. Guidelines
* Character vectors: measure, and advise best practices.
* Figures on Linux and Mono
Comments: I've been running a number of utilization cases (plots, regressions, summary) in an automated fashion against a few of the package datasets (Cars93 from MASS, nihills from DAAG, etc..). I'm looking to consistently reproduce the errors I'm getting while WebAPI/WCF hosting the R engine. 1) Sometimes 'geterrormsg' fails and the engine is reported as being in an inconsistent state. I can't find a single trigger for it as it happens at odd and random times. For example, when a parse exception is thrown for a syntax error, but the same error will not fail other times. 2) Stack limit messages. Again this seems random. Admittedly, I'm not happy with the way the C# SEXP object graph inherits from SafeHandle, but ANTS profiling has been inconclusive as to showing it to be an issue. FWIW, I think the graph should be aggregating a specific SafeHandle type, such as SafeHandleZeroOrMinusOneIsInvalid. One of the key concepts for finalizable objects is that you try to limit classes that end up being stuck in the finalizer queue to the smallest type you can. Aggregating the type addresses this concern as well as other concerns I have about object responsability (mentioned in a different comment). I'm not sure what the concern is about character vectors. Can that be elaborated a bit?