// static
void IndexBoundsBuilder::translateRegex(const RegexMatchExpression* rme,
OrderedIntervalList* oilOut, bool* exact) {
const string start = simpleRegex(rme->getString().c_str(), rme->getFlags().c_str(), exact);
// QLOG() << "regex bounds start is " << start << endl;
// Note that 'exact' is set by simpleRegex above.
if (!start.empty()) {
string end = start;
end[end.size() - 1]++;
oilOut->intervals.push_back(makeRangeInterval(start, end, true, false));
}
else {
BSONObjBuilder bob;
bob.appendMinForType("", String);
bob.appendMaxForType("", String);
BSONObj dataObj = bob.obj();
verify(dataObj.isOwned());
oilOut->intervals.push_back(makeRangeInterval(dataObj, true, false));
}
// Regexes are after strings.
BSONObjBuilder bob;
bob.appendRegex("", rme->getString(), rme->getFlags());
oilOut->intervals.push_back(makePointInterval(bob.obj()));
}
// static
void IndexBoundsBuilder::translateEquality(const BSONElement& data, bool isHashed,
OrderedIntervalList* oil, bool* exact) {
// We have to copy the data out of the parse tree and stuff it into the index
// bounds. BSONValue will be useful here.
BSONObj dataObj;
if (isHashed) {
dataObj = ExpressionMapping::hash(data);
}
else {
dataObj = objFromElement(data);
}
// UNITTEST 11738048
if (Array == dataObj.firstElement().type()) {
// XXX: bad
oil->intervals.push_back(allValues());
*exact = false;
}
else {
verify(dataObj.isOwned());
oil->intervals.push_back(makePointInterval(dataObj));
// XXX: it's exact if the index isn't sparse?
if (dataObj.firstElement().isNull() || isHashed) {
*exact = false;
}
else {
*exact = true;
}
}
}
PlanStage::StageState EnsureSortedStage::doWork(WorkingSetID* out) {
StageState stageState = child()->work(out);
if (PlanStage::ADVANCED == stageState) {
// We extract the sort key from the WSM's computed data. This must have been generated
// by a SortKeyGeneratorStage descendent in the execution tree.
WorkingSetMember* member = _ws->get(*out);
auto sortKeyComputedData =
static_cast<const SortKeyComputedData*>(member->getComputed(WSM_SORT_KEY));
BSONObj curSortKey = sortKeyComputedData->getSortKey();
invariant(!curSortKey.isEmpty());
if (!_prevSortKey.isEmpty() && !isInOrder(_prevSortKey, curSortKey)) {
// 'member' is out of order. Drop it from the result set.
_ws->free(*out);
++_specificStats.nDropped;
return PlanStage::NEED_TIME;
}
invariant(curSortKey.isOwned());
_prevSortKey = curSortKey;
return PlanStage::ADVANCED;
}
return stageState;
}
Status Collection::setValidator(OperationContext* txn, BSONObj validatorDoc) {
invariant(txn->lockState()->isCollectionLockedForMode(ns().toString(), MODE_X));
// Make owned early so that the parsed match expression refers to the owned object.
if (!validatorDoc.isOwned())
validatorDoc = validatorDoc.getOwned();
auto statusWithMatcher = parseValidator(validatorDoc);
if (!statusWithMatcher.isOK())
return statusWithMatcher.getStatus();
_details->updateValidator(txn, validatorDoc, getValidationLevel(), getValidationAction());
_validator = std::move(statusWithMatcher.getValue());
_validatorDoc = std::move(validatorDoc);
return Status::OK();
}
// static
void IndexBoundsBuilder::translate(const MatchExpression* expr, const BSONElement& elt,
OrderedIntervalList* oilOut, bool* exactOut) {
int direction = (elt.numberInt() >= 0) ? 1 : -1;
Interval interval;
bool exact = false;
oilOut->name = elt.fieldName();
bool isHashed = false;
if (mongoutils::str::equals("hashed", elt.valuestrsafe())) {
isHashed = true;
}
if (isHashed) {
verify(MatchExpression::EQ == expr->matchType()
|| MatchExpression::MATCH_IN == expr->matchType());
}
if (MatchExpression::EQ == expr->matchType()) {
const EqualityMatchExpression* node =
static_cast<const EqualityMatchExpression*>(expr);
// We have to copy the data out of the parse tree and stuff it into the index
// bounds. BSONValue will be useful here.
BSONObj dataObj;
if (isHashed) {
dataObj = ExpressionMapping::hash(node->getData());
}
else {
dataObj = objFromElement(node->getData());
}
// UNITTEST 11738048
if (Array == dataObj.firstElement().type()) {
// XXX: build better bounds
warning() << "building lazy bounds for " << expr->toString() << endl;
interval = allValues();
exact = false;
}
else {
verify(dataObj.isOwned());
interval = makePointInterval(dataObj);
// XXX: it's exact if the index isn't sparse
if (dataObj.firstElement().isNull()) {
exact = false;
}
else if (isHashed) {
exact = false;
}
else {
exact = true;
}
}
}
else if (MatchExpression::LTE == expr->matchType()) {
const LTEMatchExpression* node = static_cast<const LTEMatchExpression*>(expr);
BSONElement dataElt = node->getData();
BSONObjBuilder bob;
bob.appendMinForType("", dataElt.type());
bob.append(dataElt);
BSONObj dataObj = bob.obj();
verify(dataObj.isOwned());
interval = makeRangeInterval(dataObj, true, true);
// XXX: only exact if not (null or array)
exact = true;
}
else if (MatchExpression::LT == expr->matchType()) {
const LTMatchExpression* node = static_cast<const LTMatchExpression*>(expr);
BSONElement dataElt = node->getData();
BSONObjBuilder bob;
bob.appendMinForType("", dataElt.type());
bob.append(dataElt);
BSONObj dataObj = bob.obj();
verify(dataObj.isOwned());
interval = makeRangeInterval(dataObj, true, false);
// XXX: only exact if not (null or array)
exact = true;
}
else if (MatchExpression::GT == expr->matchType()) {
const GTMatchExpression* node = static_cast<const GTMatchExpression*>(expr);
BSONElement dataElt = node->getData();
BSONObjBuilder bob;
bob.append(node->getData());
bob.appendMaxForType("", dataElt.type());
BSONObj dataObj = bob.obj();
verify(dataObj.isOwned());
interval = makeRangeInterval(dataObj, false, true);
// XXX: only exact if not (null or array)
exact = true;
}
else if (MatchExpression::GTE == expr->matchType()) {
const GTEMatchExpression* node = static_cast<const GTEMatchExpression*>(expr);
BSONElement dataElt = node->getData();
BSONObjBuilder bob;
bob.append(dataElt);
bob.appendMaxForType("", dataElt.type());
BSONObj dataObj = bob.obj();
verify(dataObj.isOwned());
interval = makeRangeInterval(dataObj, true, true);
// XXX: only exact if not (null or array)
exact = true;
}
else if (MatchExpression::REGEX == expr->matchType()) {
warning() << "building lazy bounds for " << expr->toString() << endl;
interval = allValues();
exact = false;
}
else if (MatchExpression::MOD == expr->matchType()) {
BSONObjBuilder bob;
bob.appendMinForType("", NumberDouble);
bob.appendMaxForType("", NumberDouble);
BSONObj dataObj = bob.obj();
verify(dataObj.isOwned());
interval = makeRangeInterval(dataObj, true, true);
exact = false;
}
else if (MatchExpression::MATCH_IN == expr->matchType()) {
warning() << "building lazy bounds for " << expr->toString() << endl;
interval = allValues();
exact = false;
}
else if (MatchExpression::TYPE_OPERATOR == expr->matchType()) {
const TypeMatchExpression* tme = static_cast<const TypeMatchExpression*>(expr);
BSONObjBuilder bob;
bob.appendMinForType("", tme->getData());
bob.appendMaxForType("", tme->getData());
BSONObj dataObj = bob.obj();
verify(dataObj.isOwned());
interval = makeRangeInterval(dataObj, true, true);
exact = false;
}
else if (MatchExpression::MATCH_IN == expr->matchType()) {
warning() << "building lazy bounds for " << expr->toString() << endl;
interval = allValues();
exact = false;
}
else if (MatchExpression::GEO == expr->matchType()) {
const GeoMatchExpression* gme = static_cast<const GeoMatchExpression*>(expr);
// Can only do this for 2dsphere.
if (!mongoutils::str::equals("2dsphere", elt.valuestrsafe())) {
warning() << "Planner error trying to build geo bounds for " << elt.toString()
<< " index element.";
verify(0);
}
const S2Region& region = gme->getGeoQuery().getRegion();
ExpressionMapping::cover2dsphere(region, oilOut);
*exactOut = false;
// XXX: restructure this method
return;
}
else {
warning() << "Planner error, trying to build bounds for expr "
<< expr->toString() << endl;
verify(0);
}
if (-1 == direction) {
reverseInterval(&interval);
}
oilOut->intervals.push_back(interval);
*exactOut = exact;
}
// static
Status ParsedProjection::make(const BSONObj& spec,
const MatchExpression* const query,
ParsedProjection** out,
const MatchExpressionParser::WhereCallback& whereCallback) {
// Are we including or excluding fields? Values:
// -1 when we haven't initialized it.
// 1 when we're including
// 0 when we're excluding.
int include_exclude = -1;
// If any of these are 'true' the projection isn't covered.
bool include = true;
bool hasNonSimple = false;
bool hasDottedField = false;
bool includeID = true;
bool hasIndexKeyProjection = false;
bool wantGeoNearPoint = false;
bool wantGeoNearDistance = false;
// Until we see a positional or elemMatch operator we're normal.
ArrayOpType arrayOpType = ARRAY_OP_NORMAL;
BSONObjIterator it(spec);
while (it.more()) {
BSONElement e = it.next();
if (!e.isNumber() && !e.isBoolean()) {
hasNonSimple = true;
}
if (Object == e.type()) {
BSONObj obj = e.embeddedObject();
if (1 != obj.nFields()) {
return Status(ErrorCodes::BadValue, ">1 field in obj: " + obj.toString());
}
BSONElement e2 = obj.firstElement();
if (mongoutils::str::equals(e2.fieldName(), "$slice")) {
if (e2.isNumber()) {
// This is A-OK.
} else if (e2.type() == Array) {
BSONObj arr = e2.embeddedObject();
if (2 != arr.nFields()) {
return Status(ErrorCodes::BadValue, "$slice array wrong size");
}
BSONObjIterator it(arr);
// Skip over 'skip'.
it.next();
int limit = it.next().numberInt();
if (limit <= 0) {
return Status(ErrorCodes::BadValue, "$slice limit must be positive");
}
} else {
return Status(ErrorCodes::BadValue,
"$slice only supports numbers and [skip, limit] arrays");
}
} else if (mongoutils::str::equals(e2.fieldName(), "$elemMatch")) {
// Validate $elemMatch arguments and dependencies.
if (Object != e2.type()) {
return Status(ErrorCodes::BadValue,
"elemMatch: Invalid argument, object required.");
}
if (ARRAY_OP_POSITIONAL == arrayOpType) {
return Status(ErrorCodes::BadValue,
"Cannot specify positional operator and $elemMatch.");
}
if (mongoutils::str::contains(e.fieldName(), '.')) {
return Status(ErrorCodes::BadValue,
"Cannot use $elemMatch projection on a nested field.");
}
arrayOpType = ARRAY_OP_ELEM_MATCH;
// Create a MatchExpression for the elemMatch.
BSONObj elemMatchObj = e.wrap();
verify(elemMatchObj.isOwned());
// TODO: Is there a faster way of validating the elemMatchObj?
StatusWithMatchExpression swme =
MatchExpressionParser::parse(elemMatchObj, whereCallback);
if (!swme.isOK()) {
return swme.getStatus();
}
delete swme.getValue();
} else if (mongoutils::str::equals(e2.fieldName(), "$meta")) {
// Field for meta must be top level. We can relax this at some point.
if (mongoutils::str::contains(e.fieldName(), '.')) {
return Status(ErrorCodes::BadValue, "field for $meta cannot be nested");
}
// Make sure the argument to $meta is something we recognize.
// e.g. {x: {$meta: "textScore"}}
if (String != e2.type()) {
return Status(ErrorCodes::BadValue, "unexpected argument to $meta in proj");
}
if (e2.valuestr() != LiteParsedQuery::metaTextScore &&
e2.valuestr() != LiteParsedQuery::metaRecordId &&
e2.valuestr() != LiteParsedQuery::metaIndexKey &&
e2.valuestr() != LiteParsedQuery::metaGeoNearDistance &&
e2.valuestr() != LiteParsedQuery::metaGeoNearPoint) {
return Status(ErrorCodes::BadValue, "unsupported $meta operator: " + e2.str());
}
// This clobbers everything else.
if (e2.valuestr() == LiteParsedQuery::metaIndexKey) {
hasIndexKeyProjection = true;
} else if (e2.valuestr() == LiteParsedQuery::metaGeoNearDistance) {
wantGeoNearDistance = true;
} else if (e2.valuestr() == LiteParsedQuery::metaGeoNearPoint) {
wantGeoNearPoint = true;
}
} else {
return Status(ErrorCodes::BadValue,
string("Unsupported projection option: ") + e.toString());
}
} else if (mongoutils::str::equals(e.fieldName(), "_id") && !e.trueValue()) {
includeID = false;
} else {
// Projections of dotted fields aren't covered.
if (mongoutils::str::contains(e.fieldName(), '.')) {
hasDottedField = true;
}
// Validate input.
if (include_exclude == -1) {
// If we haven't specified an include/exclude, initialize include_exclude.
// We expect further include/excludes to match it.
include_exclude = e.trueValue();
include = !e.trueValue();
} else if (static_cast<bool>(include_exclude) != e.trueValue()) {
// Make sure that the incl./excl. matches the previous.
return Status(ErrorCodes::BadValue,
"Projection cannot have a mix of inclusion and exclusion.");
}
}
if (_isPositionalOperator(e.fieldName())) {
// Validate the positional op.
if (!e.trueValue()) {
return Status(ErrorCodes::BadValue,
"Cannot exclude array elements with the positional operator.");
}
if (ARRAY_OP_POSITIONAL == arrayOpType) {
return Status(ErrorCodes::BadValue,
"Cannot specify more than one positional proj. per query.");
}
if (ARRAY_OP_ELEM_MATCH == arrayOpType) {
return Status(ErrorCodes::BadValue,
"Cannot specify positional operator and $elemMatch.");
}
std::string after = mongoutils::str::after(e.fieldName(), ".$");
if (mongoutils::str::contains(after, ".$")) {
mongoutils::str::stream ss;
ss << "Positional projection '" << e.fieldName() << "' contains "
<< "the positional operator more than once.";
return Status(ErrorCodes::BadValue, ss);
}
std::string matchfield = mongoutils::str::before(e.fieldName(), '.');
if (!_hasPositionalOperatorMatch(query, matchfield)) {
mongoutils::str::stream ss;
ss << "Positional projection '" << e.fieldName() << "' does not "
<< "match the query document.";
return Status(ErrorCodes::BadValue, ss);
}
arrayOpType = ARRAY_OP_POSITIONAL;
}
}
// Fill out the returned obj.
unique_ptr<ParsedProjection> pp(new ParsedProjection());
// The positional operator uses the MatchDetails from the query
// expression to know which array element was matched.
pp->_requiresMatchDetails = arrayOpType == ARRAY_OP_POSITIONAL;
// Save the raw spec. It should be owned by the LiteParsedQuery.
verify(spec.isOwned());
pp->_source = spec;
pp->_returnKey = hasIndexKeyProjection;
// Dotted fields aren't covered, non-simple require match details, and as for include, "if
// we default to including then we can't use an index because we don't know what we're
// missing."
pp->_requiresDocument = include || hasNonSimple || hasDottedField;
// Add geoNear projections.
pp->_wantGeoNearPoint = wantGeoNearPoint;
pp->_wantGeoNearDistance = wantGeoNearDistance;
// If it's possible to compute the projection in a covered fashion, populate _requiredFields
// so the planner can perform projection analysis.
if (!pp->_requiresDocument) {
if (includeID) {
pp->_requiredFields.push_back("_id");
}
// The only way we could be here is if spec is only simple non-dotted-field projections.
// Therefore we can iterate over spec to get the fields required.
BSONObjIterator srcIt(spec);
while (srcIt.more()) {
BSONElement elt = srcIt.next();
// We've already handled the _id field before entering this loop.
if (includeID && mongoutils::str::equals(elt.fieldName(), "_id")) {
continue;
}
if (elt.trueValue()) {
pp->_requiredFields.push_back(elt.fieldName());
}
}
}
// returnKey clobbers everything.
if (hasIndexKeyProjection) {
pp->_requiresDocument = false;
}
*out = pp.release();
return Status::OK();
}
ProjectionExec::ProjectionExec(const BSONObj& spec,
const MatchExpression* queryExpression,
const MatchExpressionParser::WhereCallback& whereCallback)
: _include(true),
_special(false),
_source(spec),
_includeID(true),
_skip(0),
_limit(-1),
_arrayOpType(ARRAY_OP_NORMAL),
_hasNonSimple(false),
_hasDottedField(false),
_queryExpression(queryExpression),
_hasReturnKey(false) {
// Are we including or excluding fields?
// -1 when we haven't initialized it.
// 1 when we're including
// 0 when we're excluding.
int include_exclude = -1;
BSONObjIterator it(_source);
while (it.more()) {
BSONElement e = it.next();
if (!e.isNumber() && !e.isBoolean()) {
_hasNonSimple = true;
}
if (Object == e.type()) {
BSONObj obj = e.embeddedObject();
verify(1 == obj.nFields());
BSONElement e2 = obj.firstElement();
if (mongoutils::str::equals(e2.fieldName(), "$slice")) {
if (e2.isNumber()) {
int i = e2.numberInt();
if (i < 0) {
add(e.fieldName(), i, -i); // limit is now positive
} else {
add(e.fieldName(), 0, i);
}
} else {
verify(e2.type() == Array);
BSONObj arr = e2.embeddedObject();
verify(2 == arr.nFields());
BSONObjIterator it(arr);
int skip = it.next().numberInt();
int limit = it.next().numberInt();
verify(limit > 0);
add(e.fieldName(), skip, limit);
}
} else if (mongoutils::str::equals(e2.fieldName(), "$elemMatch")) {
_arrayOpType = ARRAY_OP_ELEM_MATCH;
// Create a MatchExpression for the elemMatch.
BSONObj elemMatchObj = e.wrap();
verify(elemMatchObj.isOwned());
_elemMatchObjs.push_back(elemMatchObj);
StatusWithMatchExpression swme =
MatchExpressionParser::parse(elemMatchObj, whereCallback);
verify(swme.isOK());
// And store it in _matchers.
_matchers[mongoutils::str::before(e.fieldName(), '.').c_str()] = swme.getValue();
add(e.fieldName(), true);
} else if (mongoutils::str::equals(e2.fieldName(), "$meta")) {
verify(String == e2.type());
if (e2.valuestr() == LiteParsedQuery::metaTextScore) {
_meta[e.fieldName()] = META_TEXT_SCORE;
} else if (e2.valuestr() == LiteParsedQuery::metaRecordId) {
_meta[e.fieldName()] = META_RECORDID;
} else if (e2.valuestr() == LiteParsedQuery::metaGeoNearPoint) {
_meta[e.fieldName()] = META_GEONEAR_POINT;
} else if (e2.valuestr() == LiteParsedQuery::metaGeoNearDistance) {
_meta[e.fieldName()] = META_GEONEAR_DIST;
} else if (e2.valuestr() == LiteParsedQuery::metaIndexKey) {
_hasReturnKey = true;
// The index key clobbers everything so just stop parsing here.
return;
} else {
// This shouldn't happen, should be caught by parsing.
verify(0);
}
} else {
verify(0);
}
} else if (mongoutils::str::equals(e.fieldName(), "_id") && !e.trueValue()) {
_includeID = false;
} else {
add(e.fieldName(), e.trueValue());
// Projections of dotted fields aren't covered.
if (mongoutils::str::contains(e.fieldName(), '.')) {
_hasDottedField = true;
}
// Validate input.
if (include_exclude == -1) {
// If we haven't specified an include/exclude, initialize include_exclude.
// We expect further include/excludes to match it.
include_exclude = e.trueValue();
_include = !e.trueValue();
}
}
if (mongoutils::str::contains(e.fieldName(), ".$")) {
_arrayOpType = ARRAY_OP_POSITIONAL;
}
}
}
void ClusterClientCursorImpl::queueResult(const BSONObj& obj) {
invariant(obj.isOwned());
_stash.push(obj);
}
// static
void IndexBoundsBuilder::translate(const MatchExpression* expr, const BSONElement& elt,
OrderedIntervalList* oilOut, bool* exactOut) {
oilOut->name = elt.fieldName();
bool isHashed = false;
if (mongoutils::str::equals("hashed", elt.valuestrsafe())) {
isHashed = true;
}
if (isHashed) {
verify(MatchExpression::EQ == expr->matchType()
|| MatchExpression::MATCH_IN == expr->matchType());
}
if (MatchExpression::ELEM_MATCH_VALUE == expr->matchType()) {
OrderedIntervalList acc;
bool exact;
translate(expr->getChild(0), elt, &acc, &exact);
if (!exact) {
*exactOut = false;
}
for (size_t i = 1; i < expr->numChildren(); ++i) {
OrderedIntervalList next;
translate(expr->getChild(i), elt, &next, &exact);
if (!exact) {
*exactOut = false;
}
intersectize(next, &acc);
}
for (size_t i = 0; i < acc.intervals.size(); ++i) {
oilOut->intervals.push_back(acc.intervals[i]);
}
if (!oilOut->intervals.empty()) {
std::sort(oilOut->intervals.begin(), oilOut->intervals.end(), IntervalComparison);
}
}
else if (MatchExpression::EQ == expr->matchType()) {
const EqualityMatchExpression* node = static_cast<const EqualityMatchExpression*>(expr);
translateEquality(node->getData(), isHashed, oilOut, exactOut);
}
else if (MatchExpression::LTE == expr->matchType()) {
const LTEMatchExpression* node = static_cast<const LTEMatchExpression*>(expr);
BSONElement dataElt = node->getData();
// Everything is <= MaxKey.
if (MaxKey == dataElt.type()) {
oilOut->intervals.push_back(allValues());
*exactOut = true;
return;
}
BSONObjBuilder bob;
bob.appendMinForType("", dataElt.type());
bob.appendAs(dataElt, "");
BSONObj dataObj = bob.obj();
verify(dataObj.isOwned());
oilOut->intervals.push_back(makeRangeInterval(dataObj, true, true));
// XXX: only exact if not (null or array)
*exactOut = true;
}
else if (MatchExpression::LT == expr->matchType()) {
const LTMatchExpression* node = static_cast<const LTMatchExpression*>(expr);
BSONElement dataElt = node->getData();
// Everything is <= MaxKey.
if (MaxKey == dataElt.type()) {
oilOut->intervals.push_back(allValues());
*exactOut = true;
return;
}
BSONObjBuilder bob;
bob.appendMinForType("", dataElt.type());
bob.appendAs(dataElt, "");
BSONObj dataObj = bob.obj();
verify(dataObj.isOwned());
QLOG() << "data obj is " << dataObj.toString() << endl;
oilOut->intervals.push_back(makeRangeInterval(dataObj, true, false));
// XXX: only exact if not (null or array)
*exactOut = true;
}
else if (MatchExpression::GT == expr->matchType()) {
const GTMatchExpression* node = static_cast<const GTMatchExpression*>(expr);
BSONElement dataElt = node->getData();
// Everything is > MinKey.
if (MinKey == dataElt.type()) {
oilOut->intervals.push_back(allValues());
*exactOut = true;
return;
}
BSONObjBuilder bob;
bob.appendAs(node->getData(), "");
bob.appendMaxForType("", dataElt.type());
BSONObj dataObj = bob.obj();
verify(dataObj.isOwned());
oilOut->intervals.push_back(makeRangeInterval(dataObj, false, true));
// XXX: only exact if not (null or array)
*exactOut = true;
}
else if (MatchExpression::GTE == expr->matchType()) {
const GTEMatchExpression* node = static_cast<const GTEMatchExpression*>(expr);
BSONElement dataElt = node->getData();
// Everything is >= MinKey.
if (MinKey == dataElt.type()) {
oilOut->intervals.push_back(allValues());
*exactOut = true;
return;
}
BSONObjBuilder bob;
bob.appendAs(dataElt, "");
bob.appendMaxForType("", dataElt.type());
BSONObj dataObj = bob.obj();
verify(dataObj.isOwned());
oilOut->intervals.push_back(makeRangeInterval(dataObj, true, true));
// XXX: only exact if not (null or array)
*exactOut = true;
}
else if (MatchExpression::REGEX == expr->matchType()) {
const RegexMatchExpression* rme = static_cast<const RegexMatchExpression*>(expr);
translateRegex(rme, oilOut, exactOut);
}
else if (MatchExpression::MOD == expr->matchType()) {
BSONObjBuilder bob;
bob.appendMinForType("", NumberDouble);
bob.appendMaxForType("", NumberDouble);
BSONObj dataObj = bob.obj();
verify(dataObj.isOwned());
oilOut->intervals.push_back(makeRangeInterval(dataObj, true, true));
*exactOut = false;
}
else if (MatchExpression::TYPE_OPERATOR == expr->matchType()) {
const TypeMatchExpression* tme = static_cast<const TypeMatchExpression*>(expr);
BSONObjBuilder bob;
bob.appendMinForType("", tme->getData());
bob.appendMaxForType("", tme->getData());
BSONObj dataObj = bob.obj();
verify(dataObj.isOwned());
oilOut->intervals.push_back(makeRangeInterval(dataObj, true, true));
*exactOut = false;
}
else if (MatchExpression::MATCH_IN == expr->matchType()) {
const InMatchExpression* ime = static_cast<const InMatchExpression*>(expr);
const ArrayFilterEntries& afr = ime->getData();
*exactOut = true;
// Create our various intervals.
bool thisBoundExact = false;
for (BSONElementSet::iterator it = afr.equalities().begin();
it != afr.equalities().end(); ++it) {
translateEquality(*it, isHashed, oilOut, &thisBoundExact);
if (!thisBoundExact) {
*exactOut = false;
}
}
for (size_t i = 0; i < afr.numRegexes(); ++i) {
translateRegex(afr.regex(i), oilOut, &thisBoundExact);
if (!thisBoundExact) {
*exactOut = false;
}
}
// XXX: what happens here?
if (afr.hasNull()) { }
// XXX: what happens here as well?
if (afr.hasEmptyArray()) { }
unionize(oilOut);
}
else if (MatchExpression::GEO == expr->matchType()) {
const GeoMatchExpression* gme = static_cast<const GeoMatchExpression*>(expr);
// Can only do this for 2dsphere.
if (!mongoutils::str::equals("2dsphere", elt.valuestrsafe())) {
warning() << "Planner error trying to build geo bounds for " << elt.toString()
<< " index element.";
verify(0);
}
const S2Region& region = gme->getGeoQuery().getRegion();
ExpressionMapping::cover2dsphere(region, oilOut);
*exactOut = false;
}
else {
warning() << "Planner error, trying to build bounds for expr "
<< expr->toString() << endl;
verify(0);
}
}
// static
void IndexBoundsBuilder::translate(const MatchExpression* expr, int direction,
OrderedIntervalList* oilOut, bool* exactOut) {
Interval interval;
bool exact = false;
if (expr->isLeaf()) {
if (MatchExpression::EQ == expr->matchType()) {
const EqualityMatchExpression* node = static_cast<const EqualityMatchExpression*>(expr);
// We have to copy the data out of the parse tree and stuff it into the index bounds.
// BSONValue will be useful here.
BSONObj dataObj = objFromElement(node->getData());
if (dataObj.couldBeArray()) {
// XXX: build better bounds
warning() << "building lazy bounds for " << expr->toString() << endl;
interval = allValues();
exact = false;
}
else {
verify(dataObj.isOwned());
interval = makePointInterval(dataObj);
exact = true;
}
}
else if (MatchExpression::LTE == expr->matchType()) {
const LTEMatchExpression* node = static_cast<const LTEMatchExpression*>(expr);
BSONObjBuilder bob;
bob.appendMinKey("");
bob.append(node->getData());
BSONObj dataObj = bob.obj();
verify(dataObj.isOwned());
interval = makeRangeInterval(dataObj, true, true);
exact = true;
}
else if (MatchExpression::LT == expr->matchType()) {
const LTMatchExpression* node = static_cast<const LTMatchExpression*>(expr);
BSONObjBuilder bob;
bob.appendMinKey("");
bob.append(node->getData());
BSONObj dataObj = bob.obj();
verify(dataObj.isOwned());
interval = makeRangeInterval(dataObj, true, false);
exact = true;
}
else if (MatchExpression::GT == expr->matchType()) {
const GTMatchExpression* node = static_cast<const GTMatchExpression*>(expr);
BSONObjBuilder bob;
bob.append(node->getData());
bob.appendMaxKey("");
BSONObj dataObj = bob.obj();
verify(dataObj.isOwned());
interval = makeRangeInterval(dataObj, false, true);
exact = true;
}
else if (MatchExpression::GTE == expr->matchType()) {
const GTEMatchExpression* node = static_cast<const GTEMatchExpression*>(expr);
BSONObjBuilder bob;
bob.append(node->getData());
bob.appendMaxKey("");
BSONObj dataObj = bob.obj();
verify(dataObj.isOwned());
interval = makeRangeInterval(dataObj, true, true);
exact = true;
}
else {
// XXX: build better bounds
warning() << "building lazy bounds for " << expr->toString() << endl;
interval = allValues();
exact = false;
}
}
else {
// XXX: build better bounds
verify(expr->isArray());
warning() << "building lazy bounds for " << expr->toString() << endl;
interval = allValues();
exact = false;
}
if (-1 == direction) {
reverseInterval(&interval);
}
oilOut->intervals.push_back(interval);
*exactOut = exact;
}
ProjectionExec::ProjectionExec(const BSONObj& spec,
const MatchExpression* queryExpression,
const CollatorInterface* collator,
const ExtensionsCallback& extensionsCallback)
: _include(true),
_special(false),
_source(spec),
_includeID(true),
_skip(0),
_limit(-1),
_arrayOpType(ARRAY_OP_NORMAL),
_queryExpression(queryExpression),
_hasReturnKey(false),
_collator(collator) {
// Whether we're including or excluding fields.
enum class IncludeExclude { kUninitialized, kInclude, kExclude };
IncludeExclude includeExclude = IncludeExclude::kUninitialized;
BSONObjIterator it(_source);
while (it.more()) {
BSONElement e = it.next();
if (Object == e.type()) {
BSONObj obj = e.embeddedObject();
verify(1 == obj.nFields());
BSONElement e2 = obj.firstElement();
if (mongoutils::str::equals(e2.fieldName(), "$slice")) {
if (e2.isNumber()) {
int i = e2.numberInt();
if (i < 0) {
add(e.fieldName(), i, -i); // limit is now positive
} else {
add(e.fieldName(), 0, i);
}
} else {
verify(e2.type() == Array);
BSONObj arr = e2.embeddedObject();
verify(2 == arr.nFields());
BSONObjIterator it(arr);
int skip = it.next().numberInt();
int limit = it.next().numberInt();
verify(limit > 0);
add(e.fieldName(), skip, limit);
}
} else if (mongoutils::str::equals(e2.fieldName(), "$elemMatch")) {
_arrayOpType = ARRAY_OP_ELEM_MATCH;
// Create a MatchExpression for the elemMatch.
BSONObj elemMatchObj = e.wrap();
verify(elemMatchObj.isOwned());
_elemMatchObjs.push_back(elemMatchObj);
StatusWithMatchExpression statusWithMatcher =
MatchExpressionParser::parse(elemMatchObj, extensionsCallback, _collator);
verify(statusWithMatcher.isOK());
// And store it in _matchers.
_matchers[mongoutils::str::before(e.fieldName(), '.').c_str()] =
statusWithMatcher.getValue().release();
add(e.fieldName(), true);
} else if (mongoutils::str::equals(e2.fieldName(), "$meta")) {
verify(String == e2.type());
if (e2.valuestr() == QueryRequest::metaTextScore) {
_meta[e.fieldName()] = META_TEXT_SCORE;
} else if (e2.valuestr() == QueryRequest::metaSortKey) {
_sortKeyMetaFields.push_back(e.fieldName());
_meta[_sortKeyMetaFields.back()] = META_SORT_KEY;
} else if (e2.valuestr() == QueryRequest::metaRecordId) {
_meta[e.fieldName()] = META_RECORDID;
} else if (e2.valuestr() == QueryRequest::metaGeoNearPoint) {
_meta[e.fieldName()] = META_GEONEAR_POINT;
} else if (e2.valuestr() == QueryRequest::metaGeoNearDistance) {
_meta[e.fieldName()] = META_GEONEAR_DIST;
} else if (e2.valuestr() == QueryRequest::metaIndexKey) {
_hasReturnKey = true;
} else {
// This shouldn't happen, should be caught by parsing.
verify(0);
}
} else {
verify(0);
}
} else if (mongoutils::str::equals(e.fieldName(), "_id") && !e.trueValue()) {
_includeID = false;
} else {
add(e.fieldName(), e.trueValue());
// If we haven't specified an include/exclude, initialize includeExclude.
if (includeExclude == IncludeExclude::kUninitialized) {
includeExclude =
e.trueValue() ? IncludeExclude::kInclude : IncludeExclude::kExclude;
_include = !e.trueValue();
}
}
if (mongoutils::str::contains(e.fieldName(), ".$")) {
_arrayOpType = ARRAY_OP_POSITIONAL;
}
}
}
// static
Status ParsedProjection::make(const BSONObj& spec,
const MatchExpression* const query,
ParsedProjection** out,
const ExtensionsCallback& extensionsCallback) {
// Whether we're including or excluding fields.
enum class IncludeExclude { kUninitialized, kInclude, kExclude };
IncludeExclude includeExclude = IncludeExclude::kUninitialized;
bool requiresDocument = false;
bool includeID = true;
bool hasIndexKeyProjection = false;
bool wantGeoNearPoint = false;
bool wantGeoNearDistance = false;
bool wantSortKey = false;
// Until we see a positional or elemMatch operator we're normal.
ArrayOpType arrayOpType = ARRAY_OP_NORMAL;
BSONObjIterator it(spec);
while (it.more()) {
BSONElement e = it.next();
if (Object == e.type()) {
BSONObj obj = e.embeddedObject();
if (1 != obj.nFields()) {
return Status(ErrorCodes::BadValue, ">1 field in obj: " + obj.toString());
}
BSONElement e2 = obj.firstElement();
if (mongoutils::str::equals(e2.fieldName(), "$slice")) {
if (e2.isNumber()) {
// This is A-OK.
} else if (e2.type() == Array) {
BSONObj arr = e2.embeddedObject();
if (2 != arr.nFields()) {
return Status(ErrorCodes::BadValue, "$slice array wrong size");
}
BSONObjIterator it(arr);
// Skip over 'skip'.
it.next();
int limit = it.next().numberInt();
if (limit <= 0) {
return Status(ErrorCodes::BadValue, "$slice limit must be positive");
}
} else {
return Status(ErrorCodes::BadValue,
"$slice only supports numbers and [skip, limit] arrays");
}
// Projections with $slice aren't covered.
requiresDocument = true;
} else if (mongoutils::str::equals(e2.fieldName(), "$elemMatch")) {
// Validate $elemMatch arguments and dependencies.
if (Object != e2.type()) {
return Status(ErrorCodes::BadValue,
"elemMatch: Invalid argument, object required.");
}
if (ARRAY_OP_POSITIONAL == arrayOpType) {
return Status(ErrorCodes::BadValue,
"Cannot specify positional operator and $elemMatch.");
}
if (mongoutils::str::contains(e.fieldName(), '.')) {
return Status(ErrorCodes::BadValue,
"Cannot use $elemMatch projection on a nested field.");
}
arrayOpType = ARRAY_OP_ELEM_MATCH;
// Create a MatchExpression for the elemMatch.
BSONObj elemMatchObj = e.wrap();
verify(elemMatchObj.isOwned());
// TODO: Is there a faster way of validating the elemMatchObj?
StatusWithMatchExpression statusWithMatcher =
MatchExpressionParser::parse(elemMatchObj, extensionsCallback);
if (!statusWithMatcher.isOK()) {
return statusWithMatcher.getStatus();
}
// Projections with $elemMatch aren't covered.
requiresDocument = true;
} else if (mongoutils::str::equals(e2.fieldName(), "$meta")) {
// Field for meta must be top level. We can relax this at some point.
if (mongoutils::str::contains(e.fieldName(), '.')) {
return Status(ErrorCodes::BadValue, "field for $meta cannot be nested");
}
// Make sure the argument to $meta is something we recognize.
// e.g. {x: {$meta: "textScore"}}
if (String != e2.type()) {
return Status(ErrorCodes::BadValue, "unexpected argument to $meta in proj");
}
if (e2.valuestr() != LiteParsedQuery::metaTextScore &&
e2.valuestr() != LiteParsedQuery::metaRecordId &&
e2.valuestr() != LiteParsedQuery::metaIndexKey &&
e2.valuestr() != LiteParsedQuery::metaGeoNearDistance &&
e2.valuestr() != LiteParsedQuery::metaGeoNearPoint &&
e2.valuestr() != LiteParsedQuery::metaSortKey) {
return Status(ErrorCodes::BadValue, "unsupported $meta operator: " + e2.str());
}
// This clobbers everything else.
if (e2.valuestr() == LiteParsedQuery::metaIndexKey) {
hasIndexKeyProjection = true;
} else if (e2.valuestr() == LiteParsedQuery::metaGeoNearDistance) {
wantGeoNearDistance = true;
} else if (e2.valuestr() == LiteParsedQuery::metaGeoNearPoint) {
wantGeoNearPoint = true;
} else if (e2.valuestr() == LiteParsedQuery::metaSortKey) {
wantSortKey = true;
}
// Of the $meta projections, only sortKey can be covered.
if (e2.valuestr() != LiteParsedQuery::metaSortKey) {
requiresDocument = true;
}
} else {
return Status(ErrorCodes::BadValue,
string("Unsupported projection option: ") + e.toString());
}
} else if (mongoutils::str::equals(e.fieldName(), "_id") && !e.trueValue()) {
includeID = false;
} else {
// Projections of dotted fields aren't covered.
if (mongoutils::str::contains(e.fieldName(), '.')) {
requiresDocument = true;
}
// If we haven't specified an include/exclude, initialize includeExclude. We expect
// further include/excludes to match it.
if (includeExclude == IncludeExclude::kUninitialized) {
includeExclude =
e.trueValue() ? IncludeExclude::kInclude : IncludeExclude::kExclude;
} else if ((includeExclude == IncludeExclude::kInclude && !e.trueValue()) ||
(includeExclude == IncludeExclude::kExclude && e.trueValue())) {
return Status(ErrorCodes::BadValue,
"Projection cannot have a mix of inclusion and exclusion.");
}
}
if (_isPositionalOperator(e.fieldName())) {
// Validate the positional op.
if (!e.trueValue()) {
return Status(ErrorCodes::BadValue,
"Cannot exclude array elements with the positional operator.");
}
if (ARRAY_OP_POSITIONAL == arrayOpType) {
return Status(ErrorCodes::BadValue,
"Cannot specify more than one positional proj. per query.");
}
if (ARRAY_OP_ELEM_MATCH == arrayOpType) {
return Status(ErrorCodes::BadValue,
"Cannot specify positional operator and $elemMatch.");
}
std::string after = mongoutils::str::after(e.fieldName(), ".$");
if (mongoutils::str::contains(after, ".$")) {
mongoutils::str::stream ss;
ss << "Positional projection '" << e.fieldName() << "' contains "
<< "the positional operator more than once.";
return Status(ErrorCodes::BadValue, ss);
}
std::string matchfield = mongoutils::str::before(e.fieldName(), '.');
if (!_hasPositionalOperatorMatch(query, matchfield)) {
mongoutils::str::stream ss;
ss << "Positional projection '" << e.fieldName() << "' does not "
<< "match the query document.";
return Status(ErrorCodes::BadValue, ss);
}
arrayOpType = ARRAY_OP_POSITIONAL;
}
}
// If includeExclude is uninitialized or set to exclude fields, then we can't use an index
// because we don't know what fields we're missing.
if (includeExclude == IncludeExclude::kUninitialized ||
includeExclude == IncludeExclude::kExclude) {
requiresDocument = true;
}
// Fill out the returned obj.
unique_ptr<ParsedProjection> pp(new ParsedProjection());
// The positional operator uses the MatchDetails from the query
// expression to know which array element was matched.
pp->_requiresMatchDetails = arrayOpType == ARRAY_OP_POSITIONAL;
// Save the raw spec. It should be owned by the LiteParsedQuery.
verify(spec.isOwned());
pp->_source = spec;
pp->_returnKey = hasIndexKeyProjection;
pp->_requiresDocument = requiresDocument;
// Add meta-projections.
pp->_wantGeoNearPoint = wantGeoNearPoint;
pp->_wantGeoNearDistance = wantGeoNearDistance;
pp->_wantSortKey = wantSortKey;
// If it's possible to compute the projection in a covered fashion, populate _requiredFields
// so the planner can perform projection analysis.
if (!pp->_requiresDocument) {
if (includeID) {
pp->_requiredFields.push_back("_id");
}
// The only way we could be here is if spec is only simple non-dotted-field inclusions or
// the $meta sortKey projection. Therefore we can iterate over spec to get the fields
// required.
BSONObjIterator srcIt(spec);
while (srcIt.more()) {
BSONElement elt = srcIt.next();
// We've already handled the _id field before entering this loop.
if (includeID && mongoutils::str::equals(elt.fieldName(), "_id")) {
continue;
}
// $meta sortKey should not be checked as a part of _requiredFields, since it can
// potentially produce a covered projection as long as the sort key is covered.
if (BSONType::Object == elt.type()) {
dassert(elt.Obj() == BSON("$meta"
<< "sortKey"));
continue;
}
if (elt.trueValue()) {
pp->_requiredFields.push_back(elt.fieldName());
}
}
}
// returnKey clobbers everything except for sortKey meta-projection.
if (hasIndexKeyProjection && !wantSortKey) {
pp->_requiresDocument = false;
}
*out = pp.release();
return Status::OK();
}
