avoid using EXTPTR_PTR

[kevinushey]

As it's not part of the official R API. See https://r.789695.n4.nabble.com/Possible-ABI-change-in-R-4-0-1-td4764335.html for details.

We should use R_ExternalPtrAddtr() instead.

kevinushey@cdrv:~/r/pkg/Rcpp [master]
$ rg "EXTPTR_PTR"
inst/include/Rcpp/module/class.h
170:                vec_signed_method* mets = reinterpret_cast< vec_signed_method* >( EXTPTR_PTR( method_xp ) ) ;
197:                vec_signed_method* mets = reinterpret_cast< vec_signed_method* >( EXTPTR_PTR( method_xp ) ) ;
219:                vec_signed_method* mets = reinterpret_cast< vec_signed_method* >( EXTPTR_PTR( method_xp ) ) ;
396:                prop_class* prop = reinterpret_cast< prop_class* >( EXTPTR_PTR( field_xp ) ) ;
403:                prop_class* prop = reinterpret_cast< prop_class* >( EXTPTR_PTR( field_xp ) ) ;

src/api.cpp
148:    snprintf(buffer, 20, "%p", (void*)EXTPTR_PTR(xp));

src/module.cpp
56:    return EXTPTR_PTR(xp) == 0;

Evidently this change was done way back in 2011 as well for some usages:

Rcpp/ChangeLog

Lines 6630 to 6634 in b125470

	2011-02-10 Douglas Bates <bates@stat.wisc.edu>
	* inst/include/Rcpp/XPtr.h: Replace calls to EXTPTR_PTR with
	R_ExternalPtrAddr. The distinction is important when R is
	compiled with PROTECTCHECK enabled.

[eddelbuettel]

Sure, good catch. Saw that discussion, and of course the ping-pong between @gaborcsardi and myself about the roxygen2 issue he was pinning on us which I didn't want to have here ... til @jeroen found this underlying change.

Which, almost surely, is not related to the roxygen2 issue as the above (which I also see here using ag rather than rg) affects only Modules which "some" package use, but not many -- and almost surely not roxygen2.

Let me chew over if I want to tackle this this week or rather ship what we have now as 1.0.5 and then let a change such as this "rest and mature" a little en route to 1.0.6. If that release happens. Because I am still not over the nonsense I had to put up with in the discussion after 1.0.4 came out, and we had risked late changes. Very little appetite for repeating that....

[eddelbuettel]

I mistook the reference from src/api.cpp for something else at first. That is a pretty-printer, rather than the package status, so that could indeed be called more widely.

[eddelbuettel]

By the same token this and the code below in the same file may need a rethink:

Rcpp/src/barrier.cpp

Line 24 in 97222bb

#define USE_RINTERNALS

[eddelbuettel]

Hm, and R_ExternalPtrAddr() is from R_Internals.h, which is not part of the API. There is a reference to Chap 6 of WRE but that does clarify too much either.

Ok, it is in the 'else' branch of #ifdef USE_RINTERNALS so that;s clearer. False alarm by me.