update C++ code using cpp11

[Jiefei-Wang]

The argument and return type of the C++ functions have been changed to cpp11 type.

[PeteHaitch]

What's the advantage of this over the existing code?

[Jiefei-Wang]

Hi @PeteHaitch , here are two reasons for the change

1. (main reason) the current implementation does not handle the C++ level exception properly. The boost library can throw exceptions in some special cases and, without the trycatch block, will crash the R, so we need to handle it properly.
2. Since I'm working on this part, I think it will make future development easier if we rewrite the old code using the feature provided by cpp11. Therefore, we do not have to create the trycatch block ourself and the parameters for C++ functions are not an obscure type of SEXP, but a more specific C++ type.

I hope this can convince you of the changes.

Jiefei

[PeteHaitch]

Thanks, @Jiefei-Wang! Sorry, my question came across as a bit hostile, I apologise. I was just curious to understand the motivation.

[Jiefei-Wang]

No worries @PeteHaitch, it's my fault. This pull request is being reviewed by Martin so I incorrectly assume everyone knows the context. It is better to make it more clear:)

[mtmorgan]

The example on ?ipcreset fails

> ipcreset(id, 10)
Error: Invalid input type, expected 'integer' actual 'double'

because the updated code assumes the argument is an integer, whereas the original code coerced to integer.

-        {".ipc_lock", (DL_FUNC) & ipc_lock, 1},
+int ipc_n(cpp11::integers n_sexp)
 {
-    PROTECT(n_sexp = Rf_coerceVector(n_sexp, INTSXP));
-    bool test = IS_SCALAR(n_sexp, INTSXP) && (R_NaInt != Rf_asInteger(n_sexp));
-    if (!test)
-        Rf_error("'n' cannot be coerced to integer(1) and not NA");
-    int n = INTEGER(n_sexp)[0];
-    UNPROTECT(1);
-    return n;
+    if (n_sexp.size() != 1 || cpp11::is_na(n_sexp[0]))
+        Rf_error("'n' must be integer(1) and not NA");
+    return n_sexp[0];
 }

are there other similar problems?

[Jiefei-Wang]

hmm... I noticed this issue when I wrote the code, but I thought there will be an auto type conversion. I need to investigate it more...

[Jiefei-Wang]

Problem fixed. The integer value n will convert any numeric value to an integer. I added a new unit test in test_ipcmutex.R to avoid a similar issue happening.