internal/encoding: decode json files as a single JSON value

We were currently decoding json files just like we did jsonl and ndjson,
meaning that we were allowing any of them to contain many
newline-delimited JSON values, even if the filetype was JSON,
which should mean exactly one JSON value.

This bug was hiding in plain sight in tests like vet_altdata.txtar
and vet_trailing.txtar, which both loaded jsonl files as json.
For example, the latter had the following test passing
where both json and jsonl files contained many values:

    ! exec cue vet schema.cue data-trailing-mismatch.json
    stderr 'foo: conflicting values "789" and int'
    ! exec cue vet schema.cue data-trailing-mismatch.jsonl
    stderr 'foo: conflicting values "789" and int'

This is incorrect; the json decoder should reject anything after
the top-level JSON value has been decoded, even after a newline.
It now does reject multiple JSON values, as expected:

    ! exec cue vet schema.cue data-trailing-mismatch.json
    stderr 'invalid JSON.*after top-level value'
    ! exec cue vet schema.cue data-trailing-mismatch.jsonl
    stderr 'foo: conflicting values "789" and int'

Fix up these tests to adjust their expectations.
We also fix the JSON panics in the encoding_empty.txtar test,
as the panic only exists with the JSONL decoder but not in the simpler
JSON decoder which does not use an iterator anymore.

Updates #2714.

Signed-off-by: Daniel Martí <[email protected]>
Change-Id: I7cca8f1b149640c9232b177135a0a0af06388711
Reviewed-on: https://review.gerrithub.io/c/cue-lang/cue/+/1198874
Unity-Result: CUE porcuepine <[email protected]>
TryBot-Result: CUEcueckoo <[email protected]>
Reviewed-by: Roger Peppe <[email protected]>

Author: mvdan

cmd/cue/cmd/testdata/script/encoding_empty.txtar

@@ -11,7 +11,7 @@
 exec cue export cue: empty
 cmp stdout as-cue.stdout
 ! exec cue export json: empty
-stderr '^panic: '
+stderr 'unexpected end of JSON input'
 ! exec cue export jsonl: empty
 stderr '^panic: '
 ! exec cue export yaml: empty
@@ -24,7 +24,7 @@ cmp stdout as-toml.stdout
 exec cue export cue: newlines
 cmp stdout as-cue.stdout
 ! exec cue export json: newlines
-stderr '^panic: '
+stderr 'unexpected end of JSON input'
 ! exec cue export jsonl: newlines
 stderr '^panic: '
 ! exec cue export yaml: newlines

cmd/cue/cmd/testdata/script/vet_altdata.txtar

@@ -1,9 +1,13 @@
-exec cue vet schema.cue json: foo.data
+exec cue vet schema.cue jsonl: foo.data
 ! stderr .
 
-exec cue export schema.cue json: foo.data
+exec cue export schema.cue jsonl: foo.data
 cmp stdout export-stdout
 
+# Multiple JSON values should be rejected if we're not using jsonl.
+! exec cue vet schema.cue json: foo.data
+stderr 'invalid JSON.*after top-level value'
+
 -- schema.cue --
 [string]: string
 

cmd/cue/cmd/testdata/script/vet_trailing.txtar

@@ -3,23 +3,23 @@
 
 # First, sanity check that other commands spot the trailing invalid syntax.
 ! exec cue eval   data-trailing-garbage.json
-stderr 'invalid JSON'
+stderr 'invalid JSON.*after top-level value'
 ! exec cue def    data-trailing-garbage.json
-stderr 'invalid JSON'
+stderr 'invalid JSON.*after top-level value'
 ! exec cue export data-trailing-garbage.json
-stderr 'invalid JSON'
+stderr 'invalid JSON.*after top-level value'
 ! exec cue eval   data-trailing-garbage.jsonl
-stderr 'invalid JSON'
+stderr 'invalid JSON.*looking for beginning of value'
 
 # Then, check that vet does too.
 ! exec cue vet schema.cue data-trailing-garbage.json
-stderr 'invalid JSON'
+stderr 'invalid JSON.*after top-level value'
 ! exec cue vet schema.cue data-trailing-garbage.jsonl
-stderr 'invalid JSON'
+stderr 'invalid JSON.*looking for beginning of value'
 
 # Check that even the last value is validated as well.
 ! exec cue vet schema.cue data-trailing-mismatch.json
-stderr 'foo: conflicting values "789" and int'
+stderr 'invalid JSON.*after top-level value'
 ! exec cue vet schema.cue data-trailing-mismatch.jsonl
 stderr 'foo: conflicting values "789" and int'
 

internal/encoding/encoding.go

@@ -240,7 +240,17 @@ func NewDecoder(ctx *cue.Context, f *build.File, cfg *Config) *Decoder {
 		if i.err == nil {
 			i.doInterpret()
 		}
-	case build.JSON, build.JSONL:
+	case build.JSON:
+		b, err := io.ReadAll(r)
+		if err != nil {
+			i.err = err
+			break
+		}
+		i.expr, i.err = json.Extract(path, b)
+		if i.err == nil {
+			i.doInterpret()
+		}
+	case build.JSONL:
 		i.next = json.NewDecoder(nil, path, r).Extract
 		i.Next()
 	case build.YAML: