[Rd] invisible functions
The survival package, like many others, has several helper functions that are not declared in the namespace, since their only use is to be called by other "main" functions of the package. This works well since the functions in the survival namespace can see them --- without ::: arguments --- and others don't. Until a situation I ran into this week, for which I solicit comments or advice. The concordance function is a new addition, and it has one case where the same underlying helper function is called multiple times, with many arguments passed through from the parent. I thought that this would be a good use for the trick we use for model.frame, so I have code like this: concordance.coxph <- function(fit, ..., newdata, group, ymin, ymax, timewt=c("n", "S", "S/G", "n/G", "n/G2"), influence=0, ranks=FALSE, timefix=TRUE) { Call <- match.call() . . . cargs <- c("ymin", "ymax","influence", "ranks", "timewt", "timefix") cfun <- Call[c(1, match(cargs, names(Call), nomatch=0))] cfun[[1]] <- quote(cord.work) cfun$reverse <- TRUE rval <- eval(cfun, parent.frame()) This worked fine in my not-in-a-namespace test bed, but then fails when packaged up for real: the code can't find the helper function cord.work! The rule that survival package functions can "see" their undeclared helpers fails. I got it working by changing parent.frame() to environment(concordance) in the eval() call. Since everything used by cord.work is explicitly passed in its argument list this does work. Comments or suggestions? (I avoid having survival:: in the survival package because it messes up my particular test bed.) Terry [[alternative HTML version deleted]] __ R-devel@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-devel
Re: [Rd] invisible functions
On 16/10/2018 6:42 PM, Therneau, Terry M., Ph.D. via R-devel wrote: The survival package, like many others, has several helper functions that are not declared in the namespace, since their only use is to be called by other "main" functions of the package. This works well since the functions in the survival namespace can see them --- without ::: arguments --- and others don't. Until a situation I ran into this week, for which I solicit comments or advice. The concordance function is a new addition, and it has one case where the same underlying helper function is called multiple times, with many arguments passed through from the parent. I thought that this would be a good use for the trick we use for model.frame, so I have code like this: concordance.coxph <- function(fit, ..., newdata, group, ymin, ymax, timewt=c("n", "S", "S/G", "n/G", "n/G2"), influence=0, ranks=FALSE, timefix=TRUE) { Call <- match.call() . . . cargs <- c("ymin", "ymax","influence", "ranks", "timewt", "timefix") cfun <- Call[c(1, match(cargs, names(Call), nomatch=0))] cfun[[1]] <- quote(cord.work) cfun$reverse <- TRUE rval <- eval(cfun, parent.frame()) This worked fine in my not-in-a-namespace test bed, but then fails when packaged up for real: the code can't find the helper function cord.work! The rule that survival package functions can "see" their undeclared helpers fails. The reason that fails is as follows: cfun, despite its name, is not a function. It's an unevaluated expression. You are evaluating it in parent.frame(), which is the caller's evaluation frame. That frame can't generally see the private frame for your package. Since it needs to see things supplied by the user, it needs to see parent.frame. It doesn't need to see anything in your evaluation frame other than cord.work, but it can't see that, which is your problem. I think there are at least two choices: 1. change cfun[[1]] <- quote(cord.work) to cfun[[1]] <- cord.work. This should work, but error messages may be messed up, because you'll be calling an anonymous function that is a copy of cord.work, rather than calling cord.work by name. 2. change cfun[[1]] <- quote(cord.work) to cfun[[1]] <- quote(survival:::cord.work). You say this will mess up your test bed. That suggests that your test bed is broken. This is a perfectly legal and valid solution. Duncan Murdoch I got it working by changing parent.frame() to environment(concordance) in the eval() call. Since everything used by cord.work is explicitly passed in its argument list this does work. Comments or suggestions? (I avoid having survival:: in the survival package because it messes up my particular test bed.) Terry [[alternative HTML version deleted]] __ R-devel@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-devel __ R-devel@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-devel
Re: [Rd] invisible functions
> 2. change cfun[[1]] <- quote(cord.work) to cfun[[1]] <- > quote(survival:::cord.work). You say this will mess up your test bed. > That suggests that your test bed is broken. This is a perfectly legal > and valid solution. Valid in a package, but forces code to call a loaded library version of a function rather than (say) a 'source'd user-space version that is under development. Being non-specific (ie omitting foo:::) means that test code would pick up the development version in the current user environment by default. That's handy for small mods. S Ellison *** This email and any attachments are confidential. Any use, copying or disclosure other than by the intended recipient is unauthorised. If you have received this message in error, please notify the sender immediately via +44(0)20 8943 7000 or notify postmas...@lgcgroup.com and delete this message and any copies from your computer and network. LGC Limited. Registered in England 2991879. Registered office: Queens Road, Teddington, Middlesex, TW11 0LY, UK __ R-devel@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-devel
Re: [Rd] invisible functions
On 17/10/2018 11:16 AM, S Ellison wrote: 2. change cfun[[1]] <- quote(cord.work) to cfun[[1]] <- quote(survival:::cord.work). You say this will mess up your test bed. That suggests that your test bed is broken. This is a perfectly legal and valid solution. Valid in a package, but forces code to call a loaded library version of a function rather than (say) a 'source'd user-space version that is under development. Being non-specific (ie omitting foo:::) means that test code would pick up the development version in the current user environment by default. That's handy for small mods. I generally rebuild a package after each change. In RStudio that's pretty simple: "Install and Restart" saves the current workspace, shuts down R, installs the revised package, and restarts R with the old workspace. It's all pretty quick. Duncan Murdoch __ R-devel@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-devel