tint/reader/wgsl: Better diagnostics for missing parentheses
This change required removing the `&&` splitting for `a & b && c` which never valid WGSL (right now). Fixed: tint:1658 Change-Id: Ideb9f1aa9cf9b9b1054a6fc65860106dc072a9dc Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/105820 Commit-Queue: Ben Clayton <bclayton@google.com> Kokoro: Kokoro <noreply+kokoro@google.com> Reviewed-by: David Neto <dneto@google.com> Reviewed-by: Dan Sinclair <dsinclair@chromium.org>
This commit is contained in:
parent
b6e1bc7d5d
commit
00aa7ef462
|
@ -248,7 +248,53 @@ constexpr const char* FriendlyName(BinaryOp op) {
|
|||
case BinaryOp::kModulo:
|
||||
return "modulo";
|
||||
}
|
||||
return "INVALID";
|
||||
return "<invalid>";
|
||||
}
|
||||
|
||||
/// @returns the WGSL operator of the BinaryOp
|
||||
/// @param op the BinaryOp
|
||||
constexpr const char* Operator(BinaryOp op) {
|
||||
switch (op) {
|
||||
case BinaryOp::kAnd:
|
||||
return "&";
|
||||
case BinaryOp::kOr:
|
||||
return "|";
|
||||
case BinaryOp::kXor:
|
||||
return "^";
|
||||
case BinaryOp::kLogicalAnd:
|
||||
return "&&";
|
||||
case BinaryOp::kLogicalOr:
|
||||
return "||";
|
||||
case BinaryOp::kEqual:
|
||||
return "==";
|
||||
case BinaryOp::kNotEqual:
|
||||
return "!=";
|
||||
case BinaryOp::kLessThan:
|
||||
return "<";
|
||||
case BinaryOp::kGreaterThan:
|
||||
return ">";
|
||||
case BinaryOp::kLessThanEqual:
|
||||
return "<=";
|
||||
case BinaryOp::kGreaterThanEqual:
|
||||
return ">=";
|
||||
case BinaryOp::kShiftLeft:
|
||||
return "<<";
|
||||
case BinaryOp::kShiftRight:
|
||||
return ">>";
|
||||
case BinaryOp::kAdd:
|
||||
return "+";
|
||||
case BinaryOp::kSubtract:
|
||||
return "-";
|
||||
case BinaryOp::kMultiply:
|
||||
return "*";
|
||||
case BinaryOp::kDivide:
|
||||
return "/";
|
||||
case BinaryOp::kModulo:
|
||||
return "%";
|
||||
default:
|
||||
break;
|
||||
}
|
||||
return "<invalid>";
|
||||
}
|
||||
|
||||
/// @param out the std::ostream to write to
|
||||
|
|
|
@ -2662,43 +2662,24 @@ Expect<ParserImpl::ExpressionList> ParserImpl::expect_argument_expression_list(
|
|||
Maybe<const ast::Expression*> ParserImpl::bitwise_expression_post_unary_expression(
|
||||
const ast::Expression* lhs) {
|
||||
auto& t = peek();
|
||||
if (!t.Is(Token::Type::kAnd) && !t.Is(Token::Type::kOr) && !t.Is(Token::Type::kXor)) {
|
||||
return Failure::kNoMatch;
|
||||
}
|
||||
|
||||
ast::BinaryOp op = ast::BinaryOp::kXor;
|
||||
if (t.Is(Token::Type::kAnd)) {
|
||||
switch (t.type()) {
|
||||
case Token::Type::kAnd:
|
||||
op = ast::BinaryOp::kAnd;
|
||||
} else if (t.Is(Token::Type::kOr)) {
|
||||
break;
|
||||
case Token::Type::kOr:
|
||||
op = ast::BinaryOp::kOr;
|
||||
break;
|
||||
case Token::Type::kXor:
|
||||
op = ast::BinaryOp::kXor;
|
||||
break;
|
||||
default:
|
||||
return Failure::kNoMatch;
|
||||
}
|
||||
next(); // Consume t
|
||||
|
||||
while (continue_parsing()) {
|
||||
auto& n = peek();
|
||||
// Handle the case of `a & b &&c` where `&c` is a unary_expression
|
||||
bool split = false;
|
||||
if (op == ast::BinaryOp::kAnd && n.Is(Token::Type::kAndAnd)) {
|
||||
next();
|
||||
split_token(Token::Type::kAnd, Token::Type::kAnd);
|
||||
split = true;
|
||||
}
|
||||
|
||||
if (!n.Is(t.type())) {
|
||||
if (n.Is(Token::Type::kAnd) || n.Is(Token::Type::kOr) || n.Is(Token::Type::kXor)) {
|
||||
return add_error(n.source(), std::string("mixing '") + std::string(t.to_name()) +
|
||||
"' and '" + std::string(n.to_name()) +
|
||||
"' requires parenthesis");
|
||||
}
|
||||
|
||||
return lhs;
|
||||
}
|
||||
// If forced to split an `&&` then we've already done the `next` above which consumes
|
||||
// the `&`. The type check above will always fail because we only split if already consuming
|
||||
// a `&` operator.
|
||||
if (!split) {
|
||||
next();
|
||||
}
|
||||
|
||||
auto rhs = unary_expression();
|
||||
if (rhs.errored) {
|
||||
return Failure::kErrored;
|
||||
|
@ -2709,6 +2690,10 @@ Maybe<const ast::Expression*> ParserImpl::bitwise_expression_post_unary_expressi
|
|||
}
|
||||
|
||||
lhs = create<ast::BinaryExpression>(t.source(), op, lhs, rhs.value);
|
||||
|
||||
if (!match(t.type())) {
|
||||
return lhs;
|
||||
}
|
||||
}
|
||||
return Failure::kErrored;
|
||||
}
|
||||
|
@ -2949,37 +2934,45 @@ Expect<const ast::Expression*> ParserImpl::expect_relational_expression_post_una
|
|||
}
|
||||
lhs = lhs_result.value;
|
||||
|
||||
auto& t = peek();
|
||||
if (match(Token::Type::kEqualEqual) || match(Token::Type::kGreaterThan) ||
|
||||
match(Token::Type::kGreaterThanEqual) || match(Token::Type::kLessThan) ||
|
||||
match(Token::Type::kLessThanEqual) || match(Token::Type::kNotEqual)) {
|
||||
auto& tok_op = peek();
|
||||
|
||||
ast::BinaryOp op = ast::BinaryOp::kNone;
|
||||
if (t.Is(Token::Type::kLessThan)) {
|
||||
switch (tok_op.type()) {
|
||||
case Token::Type::kLessThan:
|
||||
op = ast::BinaryOp::kLessThan;
|
||||
} else if (t.Is(Token::Type::kGreaterThan)) {
|
||||
break;
|
||||
case Token::Type::kGreaterThan:
|
||||
op = ast::BinaryOp::kGreaterThan;
|
||||
} else if (t.Is(Token::Type::kLessThanEqual)) {
|
||||
break;
|
||||
case Token::Type::kLessThanEqual:
|
||||
op = ast::BinaryOp::kLessThanEqual;
|
||||
} else if (t.Is(Token::Type::kGreaterThanEqual)) {
|
||||
break;
|
||||
case Token::Type::kGreaterThanEqual:
|
||||
op = ast::BinaryOp::kGreaterThanEqual;
|
||||
} else if (t.Is(Token::Type::kEqualEqual)) {
|
||||
break;
|
||||
case Token::Type::kEqualEqual:
|
||||
op = ast::BinaryOp::kEqual;
|
||||
} else if (t.Is(Token::Type::kNotEqual)) {
|
||||
break;
|
||||
case Token::Type::kNotEqual:
|
||||
op = ast::BinaryOp::kNotEqual;
|
||||
break;
|
||||
default:
|
||||
return lhs;
|
||||
}
|
||||
|
||||
auto& next = peek();
|
||||
next(); // consume tok_op
|
||||
|
||||
auto& tok_rhs = peek();
|
||||
auto rhs = shift_expression();
|
||||
if (rhs.errored) {
|
||||
return Failure::kErrored;
|
||||
}
|
||||
if (!rhs.matched) {
|
||||
return add_error(next, std::string("unable to parse right side of ") +
|
||||
std::string(t.to_name()) + " expression");
|
||||
return add_error(tok_rhs, std::string("unable to parse right side of ") +
|
||||
std::string(tok_op.to_name()) + " expression");
|
||||
}
|
||||
lhs = create<ast::BinaryExpression>(t.source(), op, lhs, rhs.value);
|
||||
}
|
||||
return lhs;
|
||||
|
||||
return create<ast::BinaryExpression>(tok_op.source(), op, lhs, rhs.value);
|
||||
}
|
||||
|
||||
// expression
|
||||
|
@ -2992,6 +2985,7 @@ Expect<const ast::Expression*> ParserImpl::expect_relational_expression_post_una
|
|||
//
|
||||
// Note, a `relational_expression` element was added to simplify many of the right sides
|
||||
Maybe<const ast::Expression*> ParserImpl::expression() {
|
||||
auto expr = [&]() -> Maybe<const ast::Expression*> {
|
||||
auto lhs = unary_expression();
|
||||
if (lhs.errored) {
|
||||
return Failure::kErrored;
|
||||
|
@ -3026,11 +3020,6 @@ Maybe<const ast::Expression*> ParserImpl::expression() {
|
|||
while (continue_parsing()) {
|
||||
auto& n = peek();
|
||||
if (!n.Is(t.type())) {
|
||||
if (n.Is(Token::Type::kAndAnd) || n.Is(Token::Type::kOrOr)) {
|
||||
return add_error(
|
||||
n.source(), std::string("mixing '") + std::string(t.to_name()) + "' and '" +
|
||||
std::string(n.to_name()) + "' requires parenthesis");
|
||||
}
|
||||
break;
|
||||
}
|
||||
next();
|
||||
|
@ -3048,6 +3037,23 @@ Maybe<const ast::Expression*> ParserImpl::expression() {
|
|||
}
|
||||
}
|
||||
return ret;
|
||||
}();
|
||||
|
||||
if (expr.matched) {
|
||||
// Note, expression is greedy an will consume all the operators of the same type
|
||||
// so, `a & a & a` would all be consumed above. If you see any binary operator
|
||||
// after this then it _must_ be a different one, and hence an error.
|
||||
if (auto* lhs = expr->As<ast::BinaryExpression>()) {
|
||||
if (auto& n = peek(); n.IsBinaryOperator()) {
|
||||
auto source = Source::Combine(expr->source, n.source());
|
||||
add_error(source, std::string("mixing '") + ast::Operator(lhs->op) + "' and '" +
|
||||
std::string(n.to_name()) + "' requires parenthesis");
|
||||
return Failure::kErrored;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
return expr;
|
||||
}
|
||||
|
||||
// singular_expression
|
||||
|
|
|
@ -87,28 +87,6 @@ TEST_F(ParserImplTest, BitwiseExpr_Or_Parses_Multiple) {
|
|||
ASSERT_TRUE(rel->rhs->As<ast::BoolLiteralExpression>()->value);
|
||||
}
|
||||
|
||||
TEST_F(ParserImplTest, BitwiseExpr_Or_MixedWithAnd_Invalid) {
|
||||
auto p = parser("a | b & c");
|
||||
auto lhs = p->unary_expression();
|
||||
auto e = p->bitwise_expression_post_unary_expression(lhs.value);
|
||||
EXPECT_FALSE(e.matched);
|
||||
EXPECT_TRUE(e.errored);
|
||||
EXPECT_EQ(e.value, nullptr);
|
||||
EXPECT_TRUE(p->has_error());
|
||||
EXPECT_EQ(p->error(), "1:7: mixing '|' and '&' requires parenthesis");
|
||||
}
|
||||
|
||||
TEST_F(ParserImplTest, BitwiseExpr_Or_MixedWithXor_Invalid) {
|
||||
auto p = parser("a | b ^ c");
|
||||
auto lhs = p->unary_expression();
|
||||
auto e = p->bitwise_expression_post_unary_expression(lhs.value);
|
||||
EXPECT_FALSE(e.matched);
|
||||
EXPECT_TRUE(e.errored);
|
||||
EXPECT_EQ(e.value, nullptr);
|
||||
EXPECT_TRUE(p->has_error());
|
||||
EXPECT_EQ(p->error(), "1:7: mixing '|' and '^' requires parenthesis");
|
||||
}
|
||||
|
||||
TEST_F(ParserImplTest, BitwiseExpr_Or_InvalidRHS) {
|
||||
auto p = parser("true | if (a) {}");
|
||||
auto lhs = p->unary_expression();
|
||||
|
@ -179,28 +157,6 @@ TEST_F(ParserImplTest, BitwiseExpr_Xor_Parses_Multiple) {
|
|||
ASSERT_TRUE(rel->rhs->As<ast::BoolLiteralExpression>()->value);
|
||||
}
|
||||
|
||||
TEST_F(ParserImplTest, BitwiseExpr_Xor_MixedWithOr_Invalid) {
|
||||
auto p = parser("a ^ b | c");
|
||||
auto lhs = p->unary_expression();
|
||||
auto e = p->bitwise_expression_post_unary_expression(lhs.value);
|
||||
EXPECT_FALSE(e.matched);
|
||||
EXPECT_TRUE(e.errored);
|
||||
EXPECT_EQ(e.value, nullptr);
|
||||
EXPECT_TRUE(p->has_error());
|
||||
EXPECT_EQ(p->error(), "1:7: mixing '^' and '|' requires parenthesis");
|
||||
}
|
||||
|
||||
TEST_F(ParserImplTest, BitwiseExpr_Xor_MixedWithAnd_Invalid) {
|
||||
auto p = parser("a ^ b & c");
|
||||
auto lhs = p->unary_expression();
|
||||
auto e = p->bitwise_expression_post_unary_expression(lhs.value);
|
||||
EXPECT_FALSE(e.matched);
|
||||
EXPECT_TRUE(e.errored);
|
||||
EXPECT_EQ(e.value, nullptr);
|
||||
EXPECT_TRUE(p->has_error());
|
||||
EXPECT_EQ(p->error(), "1:7: mixing '^' and '&' requires parenthesis");
|
||||
}
|
||||
|
||||
TEST_F(ParserImplTest, BitwiseExpr_Xor_InvalidRHS) {
|
||||
auto p = parser("true ^ if (a) {}");
|
||||
auto lhs = p->unary_expression();
|
||||
|
@ -275,59 +231,21 @@ TEST_F(ParserImplTest, BitwiseExpr_And_Parses_AndAnd) {
|
|||
auto p = parser("a & true &&b");
|
||||
auto lhs = p->unary_expression();
|
||||
auto e = p->bitwise_expression_post_unary_expression(lhs.value);
|
||||
// bitwise_expression_post_unary_expression returns before parsing '&&'
|
||||
|
||||
EXPECT_TRUE(e.matched);
|
||||
EXPECT_FALSE(e.errored);
|
||||
EXPECT_FALSE(p->has_error()) << p->error();
|
||||
ASSERT_NE(e.value, nullptr);
|
||||
|
||||
// lhs: (a & true)
|
||||
// rhs: &b
|
||||
// lhs: a
|
||||
// rhs: true
|
||||
ASSERT_TRUE(e->Is<ast::BinaryExpression>());
|
||||
auto* rel = e->As<ast::BinaryExpression>();
|
||||
EXPECT_EQ(ast::BinaryOp::kAnd, rel->op);
|
||||
|
||||
ASSERT_TRUE(rel->rhs->Is<ast::UnaryOpExpression>());
|
||||
auto* unary = rel->rhs->As<ast::UnaryOpExpression>();
|
||||
EXPECT_EQ(ast::UnaryOp::kAddressOf, unary->op);
|
||||
|
||||
ASSERT_TRUE(unary->expr->Is<ast::IdentifierExpression>());
|
||||
auto* ident = unary->expr->As<ast::IdentifierExpression>();
|
||||
EXPECT_EQ(ident->symbol, p->builder().Symbols().Register("b"));
|
||||
|
||||
ASSERT_TRUE(rel->lhs->Is<ast::BinaryExpression>());
|
||||
|
||||
// lhs: a
|
||||
// rhs: true
|
||||
rel = rel->lhs->As<ast::BinaryExpression>();
|
||||
|
||||
ASSERT_TRUE(rel->lhs->Is<ast::IdentifierExpression>());
|
||||
ident = rel->lhs->As<ast::IdentifierExpression>();
|
||||
EXPECT_EQ(ident->symbol, p->builder().Symbols().Register("a"));
|
||||
|
||||
ASSERT_TRUE(rel->rhs->Is<ast::BoolLiteralExpression>());
|
||||
ASSERT_TRUE(rel->rhs->As<ast::BoolLiteralExpression>()->value);
|
||||
}
|
||||
|
||||
TEST_F(ParserImplTest, BitwiseExpr_And_MixedWithOr_Invalid) {
|
||||
auto p = parser("a & b | c");
|
||||
auto lhs = p->unary_expression();
|
||||
auto e = p->bitwise_expression_post_unary_expression(lhs.value);
|
||||
EXPECT_FALSE(e.matched);
|
||||
EXPECT_TRUE(e.errored);
|
||||
EXPECT_EQ(e.value, nullptr);
|
||||
EXPECT_TRUE(p->has_error());
|
||||
EXPECT_EQ(p->error(), "1:7: mixing '&' and '|' requires parenthesis");
|
||||
}
|
||||
|
||||
TEST_F(ParserImplTest, BitwiseExpr_And_MixedWithXor_Invalid) {
|
||||
auto p = parser("a & b ^ c");
|
||||
auto lhs = p->unary_expression();
|
||||
auto e = p->bitwise_expression_post_unary_expression(lhs.value);
|
||||
EXPECT_FALSE(e.matched);
|
||||
EXPECT_TRUE(e.errored);
|
||||
EXPECT_EQ(e.value, nullptr);
|
||||
EXPECT_TRUE(p->has_error());
|
||||
EXPECT_EQ(p->error(), "1:7: mixing '&' and '^' requires parenthesis");
|
||||
}
|
||||
|
||||
TEST_F(ParserImplTest, BitwiseExpr_And_InvalidRHS) {
|
||||
|
|
|
@ -165,7 +165,7 @@ TEST_F(ParserImplTest, Expression_Mixing_OrWithAnd) {
|
|||
EXPECT_TRUE(e.errored);
|
||||
EXPECT_EQ(e.value, nullptr);
|
||||
EXPECT_TRUE(p->has_error());
|
||||
EXPECT_EQ(p->error(), "1:11: mixing '&&' and '||' requires parenthesis");
|
||||
EXPECT_EQ(p->error(), "1:3: mixing '&&' and '||' requires parenthesis");
|
||||
}
|
||||
|
||||
TEST_F(ParserImplTest, Expression_Mixing_AndWithOr) {
|
||||
|
@ -175,7 +175,7 @@ TEST_F(ParserImplTest, Expression_Mixing_AndWithOr) {
|
|||
EXPECT_TRUE(e.errored);
|
||||
EXPECT_EQ(e.value, nullptr);
|
||||
EXPECT_TRUE(p->has_error());
|
||||
EXPECT_EQ(p->error(), "1:11: mixing '||' and '&&' requires parenthesis");
|
||||
EXPECT_EQ(p->error(), "1:3: mixing '||' and '&&' requires parenthesis");
|
||||
}
|
||||
|
||||
TEST_F(ParserImplTest, Expression_Bitwise) {
|
||||
|
@ -250,5 +250,123 @@ TEST_F(ParserImplTest, Expression_InvalidRelational) {
|
|||
EXPECT_EQ(p->error(), "1:6: unable to parse right side of <= expression");
|
||||
}
|
||||
|
||||
namespace mixing_binary_ops {
|
||||
|
||||
struct BinaryOperatorInfo {
|
||||
// A uint64_t with a single bit assigned that uniquely identifies the binary-op.
|
||||
uint64_t bit;
|
||||
// The WGSL operator symbol. Example: '<='
|
||||
const char* symbol;
|
||||
// A bit mask of all operators that can immediately follow the RHS of this operator without
|
||||
// requiring parentheses. In other words, `can_follow_without_paren` is the full set of
|
||||
// operators that can substitute `<next-operator>` in the WGSL:
|
||||
// `expr_a <this-operator> expr_b <next-operator> expr_c`
|
||||
// without requiring additional parentheses.
|
||||
uint64_t can_follow_without_paren;
|
||||
};
|
||||
|
||||
// Each binary operator is given a unique bit in a uint64_t
|
||||
static constexpr uint64_t kOpMul = static_cast<uint64_t>(1) << 0;
|
||||
static constexpr uint64_t kOpDiv = static_cast<uint64_t>(1) << 1;
|
||||
static constexpr uint64_t kOpMod = static_cast<uint64_t>(1) << 2;
|
||||
static constexpr uint64_t kOpAdd = static_cast<uint64_t>(1) << 3;
|
||||
static constexpr uint64_t kOpSub = static_cast<uint64_t>(1) << 4;
|
||||
static constexpr uint64_t kOpBAnd = static_cast<uint64_t>(1) << 5;
|
||||
static constexpr uint64_t kOpBOr = static_cast<uint64_t>(1) << 6;
|
||||
static constexpr uint64_t kOpBXor = static_cast<uint64_t>(1) << 7;
|
||||
static constexpr uint64_t kOpShl = static_cast<uint64_t>(1) << 8;
|
||||
static constexpr uint64_t kOpShr = static_cast<uint64_t>(1) << 9;
|
||||
static constexpr uint64_t kOpLt = static_cast<uint64_t>(1) << 10;
|
||||
static constexpr uint64_t kOpGt = static_cast<uint64_t>(1) << 11;
|
||||
static constexpr uint64_t kOpLe = static_cast<uint64_t>(1) << 12;
|
||||
static constexpr uint64_t kOpGe = static_cast<uint64_t>(1) << 13;
|
||||
static constexpr uint64_t kOpEq = static_cast<uint64_t>(1) << 14;
|
||||
static constexpr uint64_t kOpNe = static_cast<uint64_t>(1) << 15;
|
||||
static constexpr uint64_t kOpLAnd = static_cast<uint64_t>(1) << 16;
|
||||
static constexpr uint64_t kOpLOr = static_cast<uint64_t>(1) << 17;
|
||||
|
||||
// Bit mask for the binary operator groups
|
||||
static constexpr uint64_t kMultiplicative = kOpMul | kOpDiv | kOpMod;
|
||||
static constexpr uint64_t kAdditive = kOpAdd | kOpSub;
|
||||
static constexpr uint64_t kShift = kOpShl | kOpShr;
|
||||
static constexpr uint64_t kRelational = kOpLt | kOpGt | kOpLe | kOpGe | kOpEq | kOpNe;
|
||||
static constexpr uint64_t kLogical = kOpLAnd | kOpLOr;
|
||||
|
||||
// The binary operator table
|
||||
static constexpr const BinaryOperatorInfo kBinaryOperators[] = {
|
||||
// multiplicative
|
||||
{kOpMul, "*", kLogical | kRelational | kAdditive | kMultiplicative},
|
||||
{kOpDiv, "/", kLogical | kRelational | kAdditive | kMultiplicative},
|
||||
{kOpMod, "%", kLogical | kRelational | kAdditive | kMultiplicative},
|
||||
// additive
|
||||
{kOpAdd, "+", kLogical | kRelational | kAdditive | kMultiplicative},
|
||||
{kOpSub, "-", kLogical | kRelational | kAdditive | kMultiplicative},
|
||||
// bitwise
|
||||
{kOpBAnd, "&", kOpBAnd},
|
||||
{kOpBOr, "|", kOpBOr},
|
||||
{kOpBXor, "^", kOpBXor},
|
||||
// shift
|
||||
{kOpShl, "<<", kLogical | kRelational},
|
||||
{kOpShr, ">>", kLogical | kRelational},
|
||||
// relational
|
||||
{kOpLt, "<", kLogical | kShift | kAdditive | kMultiplicative},
|
||||
{kOpGt, ">", kLogical | kShift | kAdditive | kMultiplicative},
|
||||
{kOpLe, "<=", kLogical | kShift | kAdditive | kMultiplicative},
|
||||
{kOpGe, ">=", kLogical | kShift | kAdditive | kMultiplicative},
|
||||
{kOpEq, "==", kLogical | kShift | kAdditive | kMultiplicative},
|
||||
{kOpNe, "!=", kLogical | kShift | kAdditive | kMultiplicative},
|
||||
// logical
|
||||
{kOpLAnd, "&&", kOpLAnd | kRelational | kShift | kAdditive | kMultiplicative},
|
||||
{kOpLOr, "||", kOpLOr | kRelational | kShift | kAdditive | kMultiplicative},
|
||||
};
|
||||
|
||||
struct Case {
|
||||
BinaryOperatorInfo lhs_op;
|
||||
BinaryOperatorInfo rhs_op;
|
||||
bool should_parse;
|
||||
};
|
||||
|
||||
static std::ostream& operator<<(std::ostream& o, const Case& c) {
|
||||
return o << "a " << c.lhs_op.symbol << " b " << c.rhs_op.symbol << " c ";
|
||||
}
|
||||
|
||||
static std::vector<Case> Cases() {
|
||||
std::vector<Case> out;
|
||||
for (auto& lhs_op : kBinaryOperators) {
|
||||
for (auto& rhs_op : kBinaryOperators) {
|
||||
bool should_parse = lhs_op.can_follow_without_paren & rhs_op.bit;
|
||||
out.push_back({lhs_op, rhs_op, should_parse});
|
||||
}
|
||||
}
|
||||
return out;
|
||||
}
|
||||
|
||||
using ParserImplMixedBinaryOpTest = ParserImplTestWithParam<Case>;
|
||||
|
||||
TEST_P(ParserImplMixedBinaryOpTest, Test) {
|
||||
std::stringstream wgsl;
|
||||
wgsl << GetParam();
|
||||
auto p = parser(wgsl.str());
|
||||
auto e = p->expression();
|
||||
if (GetParam().should_parse) {
|
||||
ASSERT_TRUE(e.matched) << e.errored;
|
||||
EXPECT_TRUE(e->Is<ast::BinaryExpression>());
|
||||
} else {
|
||||
EXPECT_FALSE(e.matched);
|
||||
EXPECT_TRUE(e.errored);
|
||||
EXPECT_EQ(e.value, nullptr);
|
||||
EXPECT_TRUE(p->has_error());
|
||||
std::stringstream expected;
|
||||
expected << "1:3: mixing '" << GetParam().lhs_op.symbol << "' and '"
|
||||
<< GetParam().rhs_op.symbol << "' requires parenthesis";
|
||||
EXPECT_EQ(p->error(), expected.str());
|
||||
}
|
||||
}
|
||||
INSTANTIATE_TEST_SUITE_P(ParserImplMixedBinaryOpTest,
|
||||
ParserImplMixedBinaryOpTest,
|
||||
testing::ValuesIn(Cases()));
|
||||
|
||||
} // namespace mixing_binary_ops
|
||||
|
||||
} // namespace
|
||||
} // namespace tint::reader::wgsl
|
||||
|
|
|
@ -220,7 +220,7 @@ TEST_F(ParserImplTest, Attribute_Align_ExpressionInvalid) {
|
|||
EXPECT_EQ(attr.value, nullptr);
|
||||
EXPECT_TRUE(p->has_error());
|
||||
|
||||
EXPECT_EQ(p->error(), "1:13: expected ')' for align attribute");
|
||||
EXPECT_EQ(p->error(), "1:9: mixing '+' and '<<' requires parenthesis");
|
||||
}
|
||||
|
||||
} // namespace
|
||||
|
|
|
@ -379,8 +379,35 @@ class Token {
|
|||
|
||||
/// @returns true if the token can be split during parse into component tokens
|
||||
bool IsSplittable() const {
|
||||
return Is(Token::Type::kShiftRight) || Is(Token::Type::kGreaterThanEqual) ||
|
||||
Is(Token::Type::kAndAnd) || Is(Token::Type::kMinusMinus);
|
||||
return Is(Type::kShiftRight) || Is(Type::kGreaterThanEqual) || Is(Type::kAndAnd) ||
|
||||
Is(Type::kMinusMinus);
|
||||
}
|
||||
|
||||
/// @returns true if the token is a binary operator
|
||||
bool IsBinaryOperator() const {
|
||||
switch (type_) {
|
||||
case Type::kAnd:
|
||||
case Type::kAndAnd:
|
||||
case Type::kEqualEqual:
|
||||
case Type::kForwardSlash:
|
||||
case Type::kGreaterThan:
|
||||
case Type::kGreaterThanEqual:
|
||||
case Type::kLessThan:
|
||||
case Type::kLessThanEqual:
|
||||
case Type::kMinus:
|
||||
case Type::kMod:
|
||||
case Type::kNotEqual:
|
||||
case Type::kOr:
|
||||
case Type::kOrOr:
|
||||
case Type::kPlus:
|
||||
case Type::kShiftLeft:
|
||||
case Type::kShiftRight:
|
||||
case Type::kStar:
|
||||
case Type::kXor:
|
||||
return true;
|
||||
default:
|
||||
return false;
|
||||
}
|
||||
}
|
||||
|
||||
/// @returns the source information for this token
|
||||
|
|
Loading…
Reference in New Issue