Kepler

can you give a snippet of R that demonstrates assign and back quote and the kinds of variable names that are causing errors when not escaped?
thanks!

Here are three examples that should cover most bases. The first one shows off backticks for quoting otherwise invalid object names. Off the top of my head, I think it's safe to use backticks pretty much everywhere. The second shows how to use the assign and get operators, which is slightly more robust and can handle names with backticks in them. The third shows a limitation with assign and get, which is that you need to use a temporary variable to make changes with functions like names and dim. I used a local wrapper to keep the global environment tidy. From what I've seen, weird names can get passed into the RExpression actor through the port name (try an input port that's just a number) or through a record (e.g., EML 2 Dataset passes a record with %).

Backticks and check.names = FALSE are probably easier to implement and will cover the vast majority of edge cases.

`1 weird name` <- data.frame(`+` = 1, `-` = 2, check.names = FALSE)
`1 weird name`

assign("` name", data.frame(`*` = 3, `/` = 4, check.names = FALSE))
get("` name")

assign("` name", local({
tmp <- get("` name")
names(tmp) <- c("`", "\"")
tmp}))
get("` name")

RExpression2 actor handles "1 weird name" port names. The caveat being that you have to escape (`) them when referenced in the script (i don't think there's a way around that).

i'm going to mark this as fixed since it works with the "next generation" RExpression actor.
we can't really go back and fix the 1.0 release.
i suppose we could go back to the original RExpression actor and place all the port name assignments in backticks (`). But if we're moving away from that implementation as it is, I'm going to advocate not doing that.

Ben, I don't seem to be getting the same results as you. Here's a
workflow under Kepler 1.x dev build 17250 that shows two failures, one
due to a bad port name and another due to bad column names in the
incoming record. I think putting backticks around the names will
handle nearly all problems. Anybody who uses backticks in variable
names is just asking for trouble.

http://www.nceas.ucsb.edu/~soong/kepler/RExpression%20Naming%20Problems.xml

Oliver

I was saying it works with the RExpression2 actor. The sample workflow you have uses the original RExpression actor.

I tried this sample workflow, but using RExpression2 actors instead. Here are my findings:
the "+" and "" port names for the RecordAssembler worked fine when the downstream RExpression2 actor received the record token.
-head(input) did not work with the RExpression2 actor when trying to print out debug information. My thinking is that head() gives an ill-defined data structure that the JNI layer doesn't exactly understand.

I think I'm more confused now. I can't seem to find an RExpression2 actor in the components list. I had assumed it was a replacement for the RExpression actor in the dev build, but that seems to have been wrong.

Original Bugzilla ID was 3909