Add log and warning about inserting empty values in trie.

[gui1117]

as a first step into fixing #5986

I updated doc and log so that it is clear that inserting an empty value is not supported. Test should fails with the debug_assert and in-production chain must log error when set an empty value in trie or child trie.

[bkchr]

Why not also add the proposed change for decl_storage directly? I think if we have this, we will have covered 98% of all failures.

[gui1117]

Why not also add the proposed change for decl_storage directly? I think if we have this, we will have covered 98% of all failures.

You mean the const_assert(size_of($value_type) != 0) kind ?
Because some type have no size but encode to some values for instance the following type is of size 0:

#[derive(Encode)]
enum StorageVersion {
#[codec(index = "3")]
    _3
}

[gui1117]

though maybe I can try to make a const_assert that value_type must not be ()

[bkchr]

I thought more about the following:

decl_storage! {
      Test: Vec<u8>,
      Test2: Option<u32>,
      Test3: Option<()>,
}

Will generate:

#[test]
fn test_Test_encode_is_not_empty() {
     assert!(!Test::get().encode().is_empty());
}

The Option<*> types are ignored for the tests, because we can not expect that the type inside the Option has a default value. But, we could generate a compile error for the Option<()> case directly.

I think that is the best we can do.

[bkchr]

We should probably make this a sub-heading as well to make it more prominent, otherwise it may goes down in the overall text?

[gui1117]

I added a section Warning at the end and kept this one.

[gui1117]

done: tests generation for storage without option. And compile error for () and Option<()> value types.

[gui1117]

ok not sure this PR is wanted anymore, I can still finish it until we fix the support of empty values.

[bkchr]

Yeah, not sure either.

[bkchr]

Given the conversation in Substrate, I don't think that this is still needed.