str
ITRbunNext(Client cntxt, MalBlkPtr mb, MalStkPtr stk, InstrPtr pci)
{
BATiter bi;
BAT *b;
oid *head;
bat *bid;
ValPtr tail;
(void) cntxt;
(void) mb;
head = (oid *) getArgReference(stk, pci, 0);
tail = getArgReference(stk,pci,1);
bid = (bat *) getArgReference(stk, pci, 2);
if ((b = BATdescriptor(*bid)) == NULL) {
throw(MAL, "iterator.nextChunk", INTERNAL_BAT_ACCESS);
}
*head = (BUN)*head + 1;
if (*head >= BUNlast(b)) {
*head = oid_nil;
BBPunfix(b->batCacheid);
return MAL_SUCCEED;
}
bi = bat_iterator(b);
VALinit(tail, b->ttype, BUNtail(bi, *(BUN*) head));
BBPunfix(b->batCacheid);
return MAL_SUCCEED;
}
/*
* @-
* The BUN- and BAT-stream manipulate a long handle, i.e.
* the destination variable. It assumes it has been set to
* zero as part of runtime stack initialization. Subsequently,
* it fetches a bun and returns the increment to the control
* variable. If it returns zero the control variable has been reset
* to zero and end of stream has been reached.
*/
str
ITRbunIterator(Client cntxt, MalBlkPtr mb, MalStkPtr stk, InstrPtr pci)
{
BATiter bi;
BAT *b;
oid *head;
bat *bid;
ValPtr tail;
(void) cntxt;
(void) mb;
head = getArgReference_oid(stk, pci, 0);
tail = &stk->stk[pci->argv[1]];
bid = getArgReference_bat(stk, pci, 2);
if ((b = BATdescriptor(*bid)) == NULL) {
throw(MAL, "iterator.nextChunk", INTERNAL_BAT_ACCESS);
}
if (BATcount(b) == 0) {
*head = oid_nil;
BBPunfix(b->batCacheid);
return MAL_SUCCEED;
}
*head = BUNfirst(b);
bi = bat_iterator(b);
VALinit(tail, b->ttype, BUNtail(bi, *(BUN*) head));
BBPunfix(b->batCacheid);
return MAL_SUCCEED;
}
/* variable management */
static void
stack_set(mvc *sql, int var, const char *name, sql_subtype *type, sql_rel *rel, sql_table *t, int view, int frame)
{
sql_var *v;
if (var == sql->sizevars) {
sql->sizevars <<= 1;
sql->vars = RENEW_ARRAY(sql_var,sql->vars,sql->sizevars);
}
v = sql->vars+var;
v->name = NULL;
v->value.vtype = 0;
v->rel = rel;
v->t = t;
v->view = view;
v->frame = frame;
v->type.type = NULL;
if (type) {
int tpe = type->type->localtype;
VALinit(&sql->vars[var].value, tpe, ATOMnilptr(tpe));
v->type = *type;
}
if (name)
v->name = _STRDUP(name);
}
str RAPIeval(Client cntxt, MalBlkPtr mb, MalStkPtr stk, InstrPtr pci, bit grouped) {
sql_func * sqlfun = NULL;
str exprStr = *getArgReference_str(stk, pci, pci->retc + 1);
SEXP x, env, retval;
SEXP varname = R_NilValue;
SEXP varvalue = R_NilValue;
ParseStatus status;
int i = 0;
char argbuf[64];
char *argnames = NULL;
size_t argnameslen;
size_t pos;
char* rcall = NULL;
size_t rcalllen;
int ret_cols = 0; /* int because pci->retc is int, too*/
str *args;
int evalErr;
char *msg = MAL_SUCCEED;
BAT *b;
node * argnode;
int seengrp = FALSE;
rapiClient = cntxt;
if (!RAPIEnabled()) {
throw(MAL, "rapi.eval",
"Embedded R has not been enabled. Start server with --set %s=true",
rapi_enableflag);
}
if (!rapiInitialized) {
throw(MAL, "rapi.eval",
"Embedded R initialization has failed");
}
if (!grouped) {
sql_subfunc *sqlmorefun = (*(sql_subfunc**) getArgReference(stk, pci, pci->retc));
if (sqlmorefun) sqlfun = (*(sql_subfunc**) getArgReference(stk, pci, pci->retc))->func;
} else {
sqlfun = *(sql_func**) getArgReference(stk, pci, pci->retc);
}
args = (str*) GDKzalloc(sizeof(str) * pci->argc);
if (args == NULL) {
throw(MAL, "rapi.eval", SQLSTATE(HY001) MAL_MALLOC_FAIL);
}
// get the lock even before initialization of the R interpreter, as this can take a second and must be done only once.
MT_lock_set(&rapiLock);
env = PROTECT(eval(lang1(install("new.env")), R_GlobalEnv));
assert(env != NULL);
// first argument after the return contains the pointer to the sql_func structure
// NEW macro temporarily renamed to MNEW to allow including sql_catalog.h
if (sqlfun != NULL && sqlfun->ops->cnt > 0) {
int carg = pci->retc + 2;
argnode = sqlfun->ops->h;
while (argnode) {
char* argname = ((sql_arg*) argnode->data)->name;
args[carg] = GDKstrdup(argname);
carg++;
argnode = argnode->next;
}
}
// the first unknown argument is the group, we don't really care for the rest.
argnameslen = 2;
for (i = pci->retc + 2; i < pci->argc; i++) {
if (args[i] == NULL) {
if (!seengrp && grouped) {
args[i] = GDKstrdup("aggr_group");
seengrp = TRUE;
} else {
snprintf(argbuf, sizeof(argbuf), "arg%i", i - pci->retc - 1);
args[i] = GDKstrdup(argbuf);
}
}
argnameslen += strlen(args[i]) + 2; /* extra for ", " */
}
// install the MAL variables into the R environment
// we can basically map values to int ("INTEGER") or double ("REAL")
for (i = pci->retc + 2; i < pci->argc; i++) {
int bat_type = getBatType(getArgType(mb,pci,i));
// check for BAT or scalar first, keep code left
if (!isaBatType(getArgType(mb,pci,i))) {
b = COLnew(0, getArgType(mb, pci, i), 0, TRANSIENT);
if (b == NULL) {
msg = createException(MAL, "rapi.eval", SQLSTATE(HY001) MAL_MALLOC_FAIL);
goto wrapup;
}
if ( getArgType(mb,pci,i) == TYPE_str) {
if (BUNappend(b, *getArgReference_str(stk, pci, i), false) != GDK_SUCCEED) {
BBPreclaim(b);
b = NULL;
msg = createException(MAL, "rapi.eval", SQLSTATE(HY001) MAL_MALLOC_FAIL);
goto wrapup;
}
} else {
if (BUNappend(b, getArgReference(stk, pci, i), false) != GDK_SUCCEED) {
BBPreclaim(b);
b = NULL;
msg = createException(MAL, "rapi.eval", SQLSTATE(HY001) MAL_MALLOC_FAIL);
goto wrapup;
}
}
} else {
b = BATdescriptor(*getArgReference_bat(stk, pci, i));
if (b == NULL) {
msg = createException(MAL, "rapi.eval", SQLSTATE(HY001) MAL_MALLOC_FAIL);
goto wrapup;
}
}
// check the BAT count, if it is bigger than RAPI_MAX_TUPLES, fail
if (BATcount(b) > RAPI_MAX_TUPLES) {
msg = createException(MAL, "rapi.eval",
"Got "BUNFMT" rows, but can only handle "LLFMT". Sorry.",
BATcount(b), (lng) RAPI_MAX_TUPLES);
BBPunfix(b->batCacheid);
goto wrapup;
}
varname = PROTECT(Rf_install(args[i]));
varvalue = bat_to_sexp(b, bat_type);
if (varvalue == NULL) {
msg = createException(MAL, "rapi.eval", "unknown argument type ");
goto wrapup;
}
BBPunfix(b->batCacheid);
// install vector into R environment
Rf_defineVar(varname, varvalue, env);
UNPROTECT(2);
}
/* we are going to evaluate the user function within an anonymous function call:
* ret <- (function(arg1){return(arg1*2)})(42)
* the user code is put inside the {}, this keeps our environment clean (TM) and gives
* a clear path for return values, namely using the builtin return() function
* this is also compatible with PL/R
*/
pos = 0;
argnames = malloc(argnameslen);
if (argnames == NULL) {
msg = createException(MAL, "rapi.eval", SQLSTATE(HY001) MAL_MALLOC_FAIL);
goto wrapup;
}
argnames[0] = '\0';
for (i = pci->retc + 2; i < pci->argc; i++) {
pos += snprintf(argnames + pos, argnameslen - pos, "%s%s",
args[i], i < pci->argc - 1 ? ", " : "");
}
rcalllen = 2 * pos + strlen(exprStr) + 100;
rcall = malloc(rcalllen);
if (rcall == NULL) {
msg = createException(MAL, "rapi.eval", SQLSTATE(HY001) MAL_MALLOC_FAIL);
goto wrapup;
}
snprintf(rcall, rcalllen,
"ret <- as.data.frame((function(%s){%s})(%s), nm=NA, stringsAsFactors=F)\n",
argnames, exprStr, argnames);
free(argnames);
argnames = NULL;
#ifdef _RAPI_DEBUG_
printf("# R call %s\n",rcall);
#endif
x = R_ParseVector(mkString(rcall), 1, &status, R_NilValue);
if (LENGTH(x) != 1 || status != PARSE_OK) {
msg = createException(MAL, "rapi.eval",
"Error parsing R expression '%s'. ", exprStr);
goto wrapup;
}
retval = R_tryEval(VECTOR_ELT(x, 0), env, &evalErr);
if (evalErr != FALSE) {
char* errormsg = strdup(R_curErrorBuf());
size_t c;
if (errormsg == NULL) {
msg = createException(MAL, "rapi.eval", "Error running R expression.");
goto wrapup;
}
// remove newlines from error message so it fits into a MAPI error (lol)
for (c = 0; c < strlen(errormsg); c++) {
if (errormsg[c] == '\r' || errormsg[c] == '\n') {
errormsg[c] = ' ';
}
}
msg = createException(MAL, "rapi.eval",
"Error running R expression: %s", errormsg);
free(errormsg);
goto wrapup;
}
// ret should be a data frame with exactly as many columns as we need from retc
ret_cols = LENGTH(retval);
if (ret_cols != pci->retc) {
msg = createException(MAL, "rapi.eval",
"Expected result of %d columns, got %d", pci->retc, ret_cols);
goto wrapup;
}
// collect the return values
for (i = 0; i < pci->retc; i++) {
SEXP ret_col = VECTOR_ELT(retval, i);
int bat_type = getBatType(getArgType(mb,pci,i));
if (bat_type == TYPE_any || bat_type == TYPE_void) {
getArgType(mb,pci,i) = bat_type;
msg = createException(MAL, "rapi.eval",
"Unknown return value, possibly projecting with no parameters.");
goto wrapup;
}
// hand over the vector into a BAT
b = sexp_to_bat(ret_col, bat_type);
if (b == NULL) {
msg = createException(MAL, "rapi.eval",
"Failed to convert column %i", i);
goto wrapup;
}
// bat return
if (isaBatType(getArgType(mb,pci,i))) {
*getArgReference_bat(stk, pci, i) = b->batCacheid;
} else { // single value return, only for non-grouped aggregations
BATiter li = bat_iterator(b);
if (VALinit(&stk->stk[pci->argv[i]], bat_type,
BUNtail(li, 0)) == NULL) { // TODO BUNtail here
msg = createException(MAL, "rapi.eval", SQLSTATE(HY001) MAL_MALLOC_FAIL);
goto wrapup;
}
}
msg = MAL_SUCCEED;
}
/* unprotect environment, so it will be eaten by the GC. */
UNPROTECT(1);
wrapup:
MT_lock_unset(&rapiLock);
if (argnames)
free(argnames);
if (rcall)
free(rcall);
for (i = 0; i < pci->argc; i++)
GDKfree(args[i]);
GDKfree(args);
return msg;
}
