Repository navigation
Improve creation of ParameterizedType - #2375
eamonnmcmanus merged 2 commits into
Conversation
- Reject non-generic raw types for TypeToken.getParameterized - Fix ParameterizedTypeImpl erroneously requiring owner type for types without owner
|
|
||
| // Note: Does not check if owner type of rawType is generic because this factory method | ||
| // does not support specifying owner type | ||
| if (typeVariables.length == 0) { |
There was a problem hiding this comment.
I ran #2376 against all of Google's internal tests and found one failure where someone was calling this method with a non-generic type and an empty typeArguments array. Of course then it is just the same as calling get(NonGeneric.class), but I think this case should continue to work. It's something of a historical quirk that the reflection API represents classes with no type parameters so differently from classes with some.
There were no other failures, so I think the other changes are probably safe.
This change would imply updating the @throws text.
There was a problem hiding this comment.
Thanks for testing this!
found one failure where someone was calling this method with a non-generic type and an empty
typeArgumentsarray
Was this actually a bug in the code or is there some reason why the code was doing this? The only situation I can imagine is if the typeArguments is created dynamically and the direct caller of getParameterized does not know whether the raw class it provides is generic; but even then I think it is a bit questionable requesting creation of a ParameterizedType when it is not actually known (and not verified) that the given raw type is generic.
Are you able to share a bit more information about the code where this failed, maybe in some redacted or simplified way?
Of course then it is just the same as calling
get(NonGeneric.class), but I think this case should continue to work.
Except that getParameterized actually gives you a TypeToken(ParameterizedType) where the ParameterizedType is bogus, instead of a TypeToken(Class). So I am really not sure if we should continue to support this.
We could change TypeToken.getParameterized to return a TypeToken(Class) in that case, but that could on the other hand also cause backward compatibility issues; though supporting TypeToken.getParameterized to create bogus TypeToken(ParameterizedType) is probably also not something we want to support.
There was a problem hiding this comment.
The code I was talking about is in a test, and looks something like this:
Type type = TypeToken.getParameterized(FooResponse.class).getType();
return new GsonBuilder()
.registerTypeAdapterFactory(AutoValueAdapterFactory.create())
.registerTypeAdapter(ImmutableList.class, new ImmutableListDeserializer())
.create()
.fromJson(json, type);I had not realized that TypeToken.getParameterized returns a ParameterizedType with no type arguments in this case. That is indeed a bizarre sort of beast, and I can't help feeling that this code works mostly by accident. Since it's test code, rather than production, and apparently nothing else in all of Google's code is doing this, I think it's probably OK to introduce this exception. We should make sure to mention it in the next release notes, though.
There was a problem hiding this comment.
For the record, I did find another instance, in non-test code this time. I think the code in question happened to be broken for unrelated reasons when I did my earlier test, which is why I didn't see the failure then. I think it might be worth catching this case and returning TypeToken.get(rawType), assuming the type-argument list is also empty.
There was a problem hiding this comment.
Ok, have created #2447 for that. I hope that is what you had in mind.
- Reject non-generic raw types for TypeToken.getParameterized - Fix ParameterizedTypeImpl erroneously requiring owner type for types without owner
Purpose
Improve creation and validation of
ParameterizedTypeDescription
TypeToken.getParameterizedParameterizedTypeImplerroneously requiring owner type for types without owner (mainly / only local classes?)assertThrowsChecklist
null@since $next-version$(
$next-version$is a special placeholder which is automatically replaced during release)TestCase)mvn clean verify javadoc:jarpasses without errors