[Rd] Calling C implementations of rnorm and friends

Luis Usier luis.henrique.usier at gmail.com
Fri Jul 1 13:13:01 CEST 2016


Gabriel,

Thanks for that! I guess I really should have figured that one out sooner,
huh?

I understand why that wouldn't be CRAN-compliant. But then, what *is* the
proper way to do it? Is there any way I can call unexported functions from
another package and have it accepted by CRAN?

Also, if I instead re-write the random variable generating functions, do
you have any idea of where the source code is in the stats package? As I
said above, I can't seem to find the source code for the functional forms.

Thanks,

Luis

On Thu, Jun 30, 2016 at 10:38 PM, Gabriel Becker <gmbecker at ucdavis.edu>
wrote:

> Luis,
>
> C_rnorm is a symbol but it's not exported.  This means that you *can* do
> this by using stats:::C_rnorm.
>
> That said, it's not exported, which means that it's not supported to do
> this. So your package likely would not be allowed on CRAN, for example.
>
> Best,
> ~G
> On Jun 30, 2016 2:08 PM, "Luis Usier" <luis.henrique.usier at gmail.com>
> wrote:
>
>> Hi all,
>>
>> Looking at the body for the function rnorm, I see that the body of the
>> function is:
>>
>>     .Call(C_rnorm, n, mean, sd)
>>
>> I want to implement functions that generate normal (and other) random
>> variables. Now, I understand that I can perfectly well just call the R
>> wrapper for these functions and that will be almost indistinguishable for
>> most purposes, but for whatever reason I wanted to try and call the C
>> function directly. My first instinct was to call them as:
>>
>>     .Call(C_rnorm, 1, 1, 1)
>>
>> This doesn't work because R assumes C_rnorm is an object. Looking at the
>> documentation for .Call, I try passing it in as a string:
>>
>>     .Call("C_rnorm", 1, 1, 1, PACKAGE = "stats")
>>
>> This doesn't work either. The help page links to getNativeSymbolInfo(),
>> which I can't make work either. It also refers me to the dyn.load()
>> function and the "Writing R  Extensions" manual.
>>
>> After reading and trying to digest those, I try
>>
>>     getDLLRegisteredRoutines("stats")
>>
>> which shows me all the C and Fortran functions as registered routines,
>> along with their number of parameters. I retrieve rnorm from the list and
>> pass it on to .Call, which then works fine.
>>
>> However, is there an easier way to do this? Specifically, I would like to
>> call the DLL registered routines from within functions in a package I am
>> writing. The manual tells me I should use useDynLib(). So if I added
>> useDynLib("stats") to my namespace, would that work? Could I then have a
>> function such as:
>>
>>     function(x, y, z) .Call(C_rnorm, x, y, z)
>>
>> in my package? If not, what is the proper way of calling these functions
>> from other packages? Should I use "C_rnorm" or "norm"?
>>
>> Also, I was looking for the C source code of rnorm, because I wanted to
>> understand how the function works. Looking at Winston Chang's github R
>> mirror, I found rnorm in the random.c file in the stats package. However,
>> the code I find for it:
>>
>>
>>
>> #define DEFRAND2_REAL(name) \
>> SEXP do_##name(SEXP sn, SEXP sa, SEXP sb) { \
>>     return random2(sn, sa, sb, name, REALSXP); \
>> }
>> DEFRAND2_REAL(rnorm)
>>
>>
>> Doesn't help me at all in understanding how it works. It should create a
>> function random2(sn, sa, sb, norm, REALSXP); I understand that is a
>> version
>> of the random2 function that returns a real S expression taking sn, sa and
>> sb as parameters. But how does find the actual functional form for the
>> normal distribution?
>>
>> I am asking because I would like to rewrite some of the other functions,
>> such as parameterizing rbeta by the mean and sample size rather than by
>> the
>> number of successes and failures and rgamma by the mean and total time
>> elapsed instead of the number of events. Once I understand how the C
>> source
>> code works, it would be hopefully not very difficult to reparameterize
>> them.
>>
>> Thanks,
>>
>> Luis Usier
>>
>>         [[alternative HTML version deleted]]
>>
>> ______________________________________________
>> R-devel at r-project.org mailing list
>> https://stat.ethz.ch/mailman/listinfo/r-devel
>>
>

	[[alternative HTML version deleted]]



More information about the R-devel mailing list