Add additional function for parsing traversals with [*] keys

hclsyntax/parse_traversal_test.go

@@ -4,11 +4,13 @@
 package hclsyntax
 
 import (
+	"fmt"
 	"testing"
 
 	"github.com/go-test/deep"
-	"github.com/hashicorp/hcl/v2"
 	"github.com/zclconf/go-cty/cty"
+
+	"github.com/hashicorp/hcl/v2"
 )
 
 func TestParseTraversalAbs(t *testing.T) {
@@ -208,10 +210,61 @@ func TestParseTraversalAbs(t *testing.T) {
 			},
 			1, // extra junk after traversal
 		},
+
+		{
+			"foo[*]",
+			hcl.Traversal{
+				hcl.TraverseRoot{
+					Name: "foo",
+					SrcRange: hcl.Range{
+						Start: hcl.Pos{Line: 1, Column: 1, Byte: 0},
+						End:   hcl.Pos{Line: 1, Column: 4, Byte: 3},
+					},
+				},
+				hcl.TraverseSplat{
+					SrcRange: hcl.Range{
+						Start: hcl.Pos{Line: 1, Column: 4, Byte: 3},
+						End:   hcl.Pos{Line: 1, Column: 7, Byte: 6},
+					},
+				},
+			},
+			0,
+		},
+		{
+			"foo.*", // Still not supporting this.
+			hcl.Traversal{
+				hcl.TraverseRoot{
+					Name: "foo",
+					SrcRange: hcl.Range{
+						Start: hcl.Pos{Line: 1, Column: 1, Byte: 0},
+						End:   hcl.Pos{Line: 1, Column: 4, Byte: 3},
+					},
+				},
+			},
+			1,
+		},
+		{
+			"foo[*].bar", // Run this through the unsupported function.
+			hcl.Traversal{
+				hcl.TraverseRoot{
+					Name: "foo",
+					SrcRange: hcl.Range{
+						Start: hcl.Pos{Line: 1, Column: 1, Byte: 0},
+						End:   hcl.Pos{Line: 1, Column: 4, Byte: 3},
+					},
+				},
+			},
+			1,
+		},
 	}
 
 	for _, test := range tests {
 		t.Run(test.src, func(t *testing.T) {
+			if test.src == "foo[*]" {
+				// Skip the test that introduces splat syntax.
+				t.Skip("skipping test for unsupported splat syntax")
+			}
+
 			got, diags := ParseTraversalAbs([]byte(test.src), "", hcl.Pos{Line: 1, Column: 1})
 			if len(diags) != test.diagCount {
 				for _, diag := range diags {
@@ -226,5 +279,26 @@ func TestParseTraversalAbs(t *testing.T) {
 				}
 			}
 		})
+
+		t.Run(fmt.Sprintf("partial_%s", test.src), func(t *testing.T) {
+			if test.src == "foo[*].bar" {
+				// Skip the test that's supposed to fail for splat syntax.
+				t.Skip("skipping test for unsupported splat syntax")
+			}
+
+			got, diags := ParseTraversalPartial([]byte(test.src), "", hcl.Pos{Line: 1, Column: 1})
+			if len(diags) != test.diagCount {
+				for _, diag := range diags {
+					t.Logf(" - %s", diag.Error())
+				}
+				t.Errorf("wrong number of diagnostics %d; want %d", len(diags), test.diagCount)
+			}
+
+			if diff := deep.Equal(got, test.want); diff != nil {
+				for _, problem := range diff {
+					t.Error(problem)
+				}
+			}
+		})
 	}
 }

hclsyntax/parser_traversal.go

@@ -4,15 +4,36 @@
 package hclsyntax
 
 import (
-	"github.com/hashicorp/hcl/v2"
 	"github.com/zclconf/go-cty/cty"
+
+	"github.com/hashicorp/hcl/v2"
 )
 
 // ParseTraversalAbs parses an absolute traversal that is assumed to consume
 // all of the remaining tokens in the peeker. The usual parser recovery
 // behavior is not supported here because traversals are not expected to
 // be parsed as part of a larger program.
 func (p *parser) ParseTraversalAbs() (hcl.Traversal, hcl.Diagnostics) {
+	return p.parseTraversal(false)
+}
+
+// ParseTraversalPartial parses an absolute traversal that is permitted
+// to contain splat ([*]) expressions. Only splat expressions within square
+// brackets are permitted ([*]); splat expressions within attribute names are
+// not permitted (.*).
+//
+// The meaning of partial here is that the traversal may be incomplete, in that
+// any splat expression indicates reference to a potentially unknown number of
+// elements.
+//
+// Traversals that include splats cannot be automatically traversed by HCL using
+// the TraversalAbs or TraversalRel methods. Instead, the caller must handle
+// the traversals manually.
+func (p *parser) ParseTraversalPartial() (hcl.Traversal, hcl.Diagnostics) {
+	return p.parseTraversal(true)
+}
+
+func (p *parser) parseTraversal(allowSplats bool) (hcl.Traversal, hcl.Diagnostics) {
 	var ret hcl.Traversal
 	var diags hcl.Diagnostics
 
@@ -127,6 +148,34 @@ func (p *parser) ParseTraversalAbs() (hcl.Traversal, hcl.Diagnostics) {
 					return ret, diags
 				}
 
+			case TokenStar:
+				if allowSplats {
+
+					p.Read() // Eat the star.
+					close := p.Read()
+					if close.Type != TokenCBrack {
+						diags = append(diags, &hcl.Diagnostic{
+							Severity: hcl.DiagError,
+							Summary:  "Unclosed index brackets",
+							Detail:   "Index key must be followed by a closing bracket.",
+							Subject:  &close.Range,
+							Context:  hcl.RangeBetween(open.Range, close.Range).Ptr(),
+						})
+					}
+
+					ret = append(ret, hcl.TraverseSplat{
+						SrcRange: hcl.RangeBetween(open.Range, close.Range),
+					})
+
+					if diags.HasErrors() {
+						return ret, diags
+					}
+
+					continue
+				}
+
+				// Otherwise, return the error below for the star.
+				fallthrough
 			default:
 				if next.Type == TokenStar {
 					diags = append(diags, &hcl.Diagnostic{

hclsyntax/public.go

@@ -118,6 +118,37 @@ func ParseTraversalAbs(src []byte, filename string, start hcl.Pos) (hcl.Traversa
 	return expr, diags
 }
 
+// ParseTraversalPartial matches the behavior of ParseTraversalAbs except
+// that it allows splat expressions ([*]) to appear in the traversal.
+//
+// The returned traversals are "partial" in that the splat expression indicates
+// an unknown value for the index.
+//
+// Traversals that include splats cannot be automatically traversed by HCL using
+// the TraversalAbs or TraversalRel methods. Instead, the caller must handle
+// the traversals manually.
+func ParseTraversalPartial(src []byte, filename string, start hcl.Pos) (hcl.Traversal, hcl.Diagnostics) {
+	tokens, diags := LexExpression(src, filename, start)
+	peeker := newPeeker(tokens, false)
+	parser := &parser{peeker: peeker}
+
+	// Bare traverals are always parsed in  "ignore newlines" mode, as if
+	// they were wrapped in parentheses.
+	parser.PushIncludeNewlines(false)
+
+	expr, parseDiags := parser.ParseTraversalPartial()
+	diags = append(diags, parseDiags...)
+
+	parser.PopIncludeNewlines()
+
+	// Panic if the parser uses incorrect stack discipline with the peeker's
+	// newlines stack, since otherwise it will produce confusing downstream
+	// errors.
+	peeker.AssertEmptyIncludeNewlinesStack()
+
+	return expr, diags
+}
+
 // LexConfig performs lexical analysis on the given buffer, treating it as a
 // whole HCL config file, and returns the resulting tokens.
 //