YAML file with no content throws `MismatchedInputException` when binding to Object type (POJO etc) -- add `YAMLReadFeature.EMPTY_DOCUMENT_AS_EMPTY_OBJECT`

[serandel]

A curious behaviour I've encountered adding Jackson to a Kotlin project...

I'm trying to deserialize a YAML file to a Kotlin data class

data class Settings(val wireframe: Boolean = false)

with the following code

mapper = ObjectMapper(YAMLFactory())
mapper.registerModule(KotlinModule()) 

val settings = settingsFileInputStream.use {
   mapper.readValue(it, Settings::class.java)
}

If the file content is

wireframe: true

it works.

But if I comment out the settings like this:

#wireframe: true

then I get the following exception:

com.fasterxml.jackson.databind.exc.MismatchedInputException: No content to map due to end-of-input

I believed it was because default parameters in the data class weren't accepted, but no. To prove that, I modified the settings like this:

data class Settings(val wireframe: Boolean = false, val foo: String = "bar")

and I can deserialize from this file:

#wireframe: true
foo: foobar

but I get the exception again if I use

#wireframe: true
#foo: foobar

Jackson version: 2.9.10 (latest stable, AFAIK)

[cowtowncoder]

Since this is YAML-specific (but not necessarily Kotlin-specific), I'll transfer to jackson-dataformats-yaml.

Problem itself is probably since what you have is essentially an empty document, and I don't know what it looks like at tokenization level -- since no properties exist, there is nothing to infer an "Object" value, and this is difficult to model.

[michaelhixson]

Here's how I worked around this issue in a Java project, in case it helps anyone.

Workaround 1

If my YAML file would be empty except for comments, add a {} at the top level to prevent the mapper from blowing up.

# This is a comment.
# foo=bar

# This {} is necessary because everything else is commented out.
{}

This "worked" but felt bad. I felt compelled to explain the {} in my example YAML config file with a comment, but I would have preferred to not distract the reader with that stuff at all. I'm no longer using this approach.

Workaround 2

Do a two-step deserialization (1) from YAML to JsonNode then (2) from JsonNode to an instance of my class. If the JsonNode is empty, which it will be when the YAML file contains only comments and white space, return a default value. Otherwise, ask the ObjectMapper to read the JsonNode.

Path yamlFile = ...
YAMLMapper yamlYapper = new YAMLMapper();
JsonNode tree;
try (InputStream inputStream = Files.newInputStream(yamlFile)) {
  tree = yamlMapper.readTree(inputStream);
}

if (tree.isEmpty()) {
  return MyConfig.defaultConfig();
}

ObjectMapper objectMapper = new ObjectMapper();
return objectMapper.treeToValue(tree, MyConfig.class);

This is the approach that I'm using currently.

[cowtowncoder]

@michaelhixson Thank you for sharing this! This makes sense as Tree Model can read "no content" as MissingNode. There are 2 other similar approaches now that you mention this, either:

1. create parser from YAMLFactory (via mapper.getFactory()), check if (parser.nextToken() == null) { // no content
2. create MappingIterator<MyConfig> via mapper.readerFor(MyConfig.class).readValues()), and see if if (hasNext()) { ... has value ... } else { ... no value create default ... }

in fact (2) might be the cleanest version:

try (MappingIterator<MyConfig> iterator = mapper.readerFor(MyConfig.class)
    .readValues()) {
  if (!iterator.hasValue()) {
    return MyConfig.defaultConfig();
  }
  return iterator.nextValue();
}

[mikebell90]

@cowtowncoder This appears to be fixed in Jackson 2.18.3 - which I updated to from 2.18.2. This test below started failing for me ;) I assume this is WAD?

    // An empty manifest throws up on the Node
    @Test
    public void testEmptyManifest() throws Exception {
        assertThrows(MismatchedInputException.class, () -> {
            try (InputStream is = new ByteArrayInputStream(new byte[0])) {
                deserialize(is);
            }
        });
    }
   private Manifest deserialize(InputStream is) throws IOException {
        JsonNode node = mapper.readTree(is);
        return objectMapper.treeToValue(node, Manifest.class);
    }

[cowtowncoder]

@mikebell90 2.18.3 should have no changes over 2.18.2 or 2.18.1 -- but 2.18.1 did upgrade SnakeYAML to 2.3. That seems like the only YAML-specific change.

But as to failure: I think it should work (if I read this right, considering Manifest def is missing).
Although as per my earlier comment, empty docs are tricky to properly represent.

What exception do you get?

[mikebell90]

@cowtowncoder well that's where I may have put my foot in my mouth ;)

It's

com.fasterxml.jackson.databind.exc.InvalidTypeIdException: Could not resolve subtype of [simple type, class com.opentable.credentials.synchronizer.manifests.Manifest]: missing type id property 'kind'
 at [Source: UNKNOWN; byte offset: #UNKNOWN]

	at com.fasterxml.jackson.databind.exc.InvalidTypeIdException.from(InvalidTypeIdException.java:43)
	at com.fasterxml.jackson.databind.DeserializationContext.missingTypeIdException(DeserializationContext.java:2051)

Manifest is actually a complicated polymorphic beast, but here's the first part

@JsonTypeInfo(use = JsonTypeInfo.Id.NAME, include = As.EXISTING_PROPERTY, property = "kind")
@JsonSubTypes({
  @JsonSubTypes.Type(value= ImmutableServiceManifest.class, name = "service"),
  @JsonSubTypes.Type(value= ImmutableSharedManifest.class, name = "shared"),
  @JsonSubTypes.Type(value= ImmutableDatabaseManifest.class, name = "database"),
  @JsonSubTypes.Type(value= ImmutableDBManifest.class, name = "db"),
  @JsonSubTypes.Type(value= ImmutablePKIDatabaseManifest.class, name = "pki"),
  @JsonSubTypes.Type(value= ImmutablePolicyManifest.class, name = "policy"),
})
public interface Manifest {
    ManifestKind getKind(); (many more properties)

So on the one hand maybe unrellated to above, though it "looked" similar. OTOH, why would this stop throwing an exception going from 2.18.2 => 2.18.3 (snakeyaml is 2.3 for both in our dep tree)

[cowtowncoder]

Yeah, this exception would come from even regular "empty" Document since -- as message states, type id kind is not included. And there is no defaultImpl defined (if there was that type would be used).

As to 2.18.3 stopping failure... jackson-databind has 11 fixes so maybe one of these changed things; polymorphic handling f.ex is fully by databind:

https://github.com/FasterXML/jackson/wiki/Jackson-Release-2.18.3

although none seems like immediately suspicious.