helper/schema: can't track changes to individual fields in a schema.Set about terraform-plugin-sdk HOT 8 CLOSED

@jbardin in hashicorp/terraform#3018 we added a default hashing function for sets that always includes all of the attributes. Do you think using that would solve the problem here?

We didn't go back and switch all the existing set users to use this default implementation because doing so requires migrating the existing state data to the new hashes in order to avoid spurious diffs after a Terraform upgrade. However, it is possible to do this and we've done it at least once. (It's also necessary to migrate each time a new attribute is added even after switching to the default implementation, since of course that too will change the hash... but I think we've not always bothered with this in cases where the diff doesn't force a new resource.)

(Sorry if I'm misunderstanding what you're reporting here.)

from terraform-plugin-sdk.

@apparentlymart: I ended up here after patching a crash in that default hashing function ;)

My first instinct was to use the default to hash all fields, but that doesn't work with any computed fields. If the hash of the config and the hash from the read phase don't match, they are considered 2 different values at 2 different paths. I started writing some special-case code to merge the difference for the common instance where a set can only contain 1 value, but that got really ugly and it would be nice to fix the complete set case like the ebs_block_device example.

from terraform-plugin-sdk.

Computed fields always pose an interesting challenge for sets, indeed... Terraform core tries to work around this in a few places, but I guess it's not perfect.

If that crash you found hasn't already been fixed it'd be nice to know more about it so we can fix it, since it's not supposed to crash. :)

from terraform-plugin-sdk.

@apparentlymart, sorry forgot to link it -- serialization fixed in hashicorp/terraform#7091

from terraform-plugin-sdk.

Adding only the fields mutable within the config to the hash value also doesn't work, because the new hash value still creates a new set in the diff, which will trigger the ForceNew for all the remaining fields.

Hi @jbardin

Why does the new hash is different from the current hash given the mutable fields are not modified?

Plus, does this issue still existed in current sdk?

from terraform-plugin-sdk.

I don't think this is still an issue and not sure what work needs to be done as its pretty old/dated. Are you experiencing a specific issue in sets/hashing?

from terraform-plugin-sdk.

Nope, this is long out of date, and I had since forgotten it was still open. We now have a defined behavior in cty.Set, and should base sdk changes of of that.

from terraform-plugin-sdk.

I'm going to lock this issue because it has been closed for 30 days ⏳. This helps our maintainers find and focus on the active issues.

If you have found a problem that seems similar to this, please open a new issue and complete the issue template so we can capture all the details necessary to investigate further.

from terraform-plugin-sdk.