Unable to deserialize stringref-enabled CBOR with ignored properties

[morokosi]

When deserializing CBOR with string references enabled using jackson-dataformats-binary, if the original string being referenced is in a field ignored by an annotation such as @JsonIgnoreProperties, the string reference cannot be resolved, causing an exception.

version: 2.17.1

record APub(String a, String b) {}
@JsonIgnoreProperties(ignoreUnknown = true)
record ASub(String b) {}

@Test
public void testCborDecode() throws IOException {
  var mapper = CBORMapper.builder.enable(CBORGenerator.Feature.STRINGREF).build();
  var aPub = new APub("foo", "foo"); // second occurrence of `foo` will be a stringref
  var aSer = mapper.writeValueAsBytes(aPub);
  var aSub = mapper.readValue(aSer, ASub.class); // <- throws com.fasterxml.jackson.core.JsonParseException: String reference (0) out of range:
}

[cowtowncoder]

Sounds like a bug indeed.

/cc @here-abarany

[cowtowncoder]

Ideally would be fixed in 2.18 branch if safe enough (that's our first Long-Term Support branch for backporting).

[cowtowncoder]

Added test

com.fasterxml.jackson.dataformat.cbor.tofix.StringRef599Test

Test added in 2.19 branch (alas, 2.18 uses old JUnit4 and has no @JacksonTestFailureExpected) -- fix itself can be backported, I hope.

[here-abarany]

My guess as to what's happening is the parser is skipping the fields entirely, which prevents any new strings from being registered for references. If the stringref extension is enabled on read, any byte or text string field will need to be parsed and registered rather than simply skipped.

[cowtowncoder]

@here-abarany CBORParser will still iterate over tokens, regardless. But there may be some optimization for case of "nextToken()" called without accessing String value of preceding JsonToken.VALUE_STRING.

[here-abarany]

Double checking the parser, there are multiple blocks such as this one:

        // For longer tokens (text, binary), we'll only read when requested
        if (_tokenIncomplete) {
            _skipIncomplete();
        }

I think what will need to be done here is to modify _skipIncomplete() and call _finishShortText()/_finishLongText() as appropriate when:

1. The stringref stack is non-empty.
2. The string isn't chunked. (as chunked strings aren't added to the reference table)