Currently, a change in the tf_vars property does not trigger a recreation but a modification of the template.

As the tf_vars effectively change the content of the template, it should trigger a new version.

As an example, it would be possible to change the name of a parameter via tf_var.

A more drastic example to show what would be possible is to base the entire coder parameters on a tf_var:

template.tf

resource "coderd_template" "example" {
  name        = "example"
  description = "example"

  versions = [
    {
      directory = "./template"
      active    = true
      tf_vars = [{
        name = "parameters"
        value = jsonencode([
          "cpu",
          "memory"
        ])
      }]
    }
  ]
}

in the template:

variable "parameters" {
  type = string
}

locals {
  parameters = jsondecode(var.parameters)
}

data "coder_parameter" "parameters" {
  for_each = toset(local.parameters)
  name     = each.key
}

If I change the content of the parameters var via the coderd_template in this scenario, the whole template would change.