Don't flatten outputs to top level for InputList and InputMap #449
Conversation
Frassle
changed the title
Frassle
changed the title
Frassle
changed the title
| @@ -46,18 +46,30 @@ namespace Pulumi | |||
| /// </summary> | |||
| public sealed class InputList<T> : Input<ImmutableArray<T>>, IEnumerable, IAsyncEnumerable<Input<T>> | |||
There was a problem hiding this comment.
Reckon it's worth extending the doc comment to include some of the awesome context you put in the PR description? Or perhaps using that as a comment for the Value property/_inputValue instance variable if you think it's better kept private/internal?
There was a problem hiding this comment.
On Value for sure, not sure it can be sensibly added to the user level description.
| var (first, second) = t; | ||
| return first.AddRange(second); | ||
| }); | ||
| return list; |
There was a problem hiding this comment.
I'm guessing Output.Concat does an Apply internally, so this use of Apply doesn't change "knownness" or anything like that?
There was a problem hiding this comment.
Yeh, if we just Concat'd the whole thing we'd end up lifting all the sub-element secret/known bits to the top level.
| } | ||
|
|
||
| Input<ImmutableDictionary<string, Input<V>>> _inputValue; | ||
| Input<ImmutableDictionary<string, Input<V>>> Value |
There was a problem hiding this comment.
Same comments as for list around doc comments/use of Apply
Frassle
force-pushed
the
fraser/inputListMap
branch
from
05e25ff to
e82f9b7
Compare
|
This PR has been shipped in release v3.73.0. |
Fixes #22
InputList and InputMap are types used to make it easier to work with lists/maps where both the overall structure or individual elements can be
Inputs.The current implementation just flattens everything to a top-level input. That is
InputList<T>is justInput<ImmutableArray<T>>. So if you start with a known empty list then add an unknown input element to it you end up with an unknown list structure. This is not ideal for previews (see #22) and doesn't align with the other SDKs where if you add an unknown element to a list the overall list shape is still known and sent to the engine for preview correctly.This PR fixes the above by keeping track of non-flattened data inside InputMap/List. i.e. the internal shape of
InputList<T>is nowInput<ImmutableArray<Input<T>>>. This allows us to take a known list and add an unknown element to it and leave the overall shape of the list still known.We do this without changing the public surface of InputList/Map, so they still inherit from
Input<ImmutableArray<T>>. We add another field for the nested structure and ensure that we just always update both fields (the one in InputList/Map and the one in the base Input class) together. In the property serializer we then use reflection to pull out the fully nested value (e.g.Input<ImmutableArray<Input<T>>>) and then recursively serialize that, rather than hitting the standardIInputcase and getting the flattened value (e.g.Input<ImmutableArray<T>>).