Previously the supported JSON plan and state formats included only
serialized output values, which was a lossy serialization of the
Terraform type system. This commit adds a type field in the usual cty
JSON format, which allows reconstitution of the original value.
For example, previously a list(string) and a set(string) containing the
same values were indistinguishable. This change serializes these as
follows:
{
"value": ["a","b","c"],
"type": ["list","string"]
}
and:
{
"value": ["a","b","c"],
"type": ["set","string"]
}
For non-interactive contexts, Terraform is typically executed with the flag -input=false.
However for runs that are not set to auto approve, the cloud integration will prompt a user for
approval input even with input being set to false. This commit enables the cloud integration to know
the value of the input flag and use it to determine whether or not to ask the user for input.
If -input is set to false and the run cannot be auto approved, the cloud integration will throw an error
stating run confirmation can no longer be handled in the CLI and that they must do so through the browser.
The plan graph does not contain all the information necessary to detect
cycles which may happen when building the apply graph. Once we have more
information from the plan we can build the complete apply graph with all
individual instances to verify that the apply can begin without errors.
The raw plan output changes were stored in the output exec node, when
they should have instead been fetch lazily through the context via the
synchronized ChangesSync value.
We had intended these functions to attempt to convert any given value, but
there is a special behavior in the function system where functions must
opt in to being able to handle dynamically-typed arguments so that we
don't need to repeat the special case for that inside every function
implementation.
In this case we _do_ want to specially handle dynamically-typed values,
because the keyword "null" in HCL produces
cty.NullVal(cty.DynamicPseudoType) and we want the conversion function
to convert it to a null of a more specific type.
These conversion functions are already just a thin wrapper around the
underlying type conversion functionality anyway, and that already supports
converting dynamic-typed values in the expected way, so we can just opt
in to allowing dynamically-typed values and let the conversion
functionality do the expected work.
Fixing this allows module authors to use type conversion functions to
give additional type information to Terraform in situations that are too
ambiguous to be handled automatically by the type inference/unification
process. Previously tostring(null) was effectively a no-op, totally
ignoring the author's request to treat the null as a string.
replace_triggered_by references are scoped to the current module, so we
need to filter changes for the current module instance. Rather than
creating a ConfigResource and filtering the result, make a
Changes.InstancesForAbsResource method to get only the AbsResource
changes.
Check for triggered resource replacement in the plan. While the
functionality of the feature works here, we ill want to follow up with a
way to indicate in the plan _why_ the resource was replaced.
The EvalContext is the only place with all the information to be able to
complete the evaluation of the replace_triggered_by expressions. These
need to be evaluated into a reference, which is then looked up in the
pending changes which the context has access too. On top of needing the
plan changes, we also need access to all providers and schemas to decode
the changes if we need to traverse the resource values for individual
attributes.
* internal/getproviders: Add URL to error message for clarity
Occasionally `terraform init` on some providers may return the following error message:
Error while installing citrix/citrixadc v1.13.0: could not query provider
registry for registry.terraform.io/citrix/citrixadc: failed to retrieve
authentication checksums for provider: 403 Forbidden
The 403 is most often returned from GitHub (rather than Registry API)
and this change makes it more obvious.
* Use Host instead of full URL
Hyphen characters are allowed in environment variable names, but are not valid POSIX variable names. Usually, it's still possible to set variable names with hyphens using utilities like env or docker. But, as a fallback, host names may encode their hyphens as double underscores in the variable name. For the example "café.fr", the variable name "TF_TOKEN_xn____caf__dma_fr" or "TF_TOKEN_xn--caf-dma_fr"
may be used.
Introduces a new method of configuring token service credentials using a host-specific environment variable. This configuration was previously possible using the [terraform-credentials-env](https://github.com/apparentlymart/terraform-credentials-env) credentials helper.
This new method is now consulted first, as it is seen to be the most proximate source of credentials before CLI configuration while still falling back to the credentials helper.
A previous change added missing quoting around object keys which do not
parse as barewords. At the same time we introduced a bug where map keys
could be double-quoted, due to calling the `displayAttributeName` helper
function (to quote non-bareword keys) then using the `writeValue` method
(which quotes all strings).
This commit fixes this and adds test coverage for map keys which require
quoting.
Data sources should not require reading the previous versions. While we
previously skipped the decoding if it were to fail, this removes the
need for any prior state at all.
The only place where the prior state was functionally used was in the
destroy path. Because a data source destroy is only for cleanup purposes
to clean out the state using the same code paths as a managed resource,
we can substitute the prior state in the change change with a null value
to maintain the same behavior.
After data source handling was moved from a separate refresh phase into
the planning phase, reading the existing state was only used for
informational purposes. This had been reduced to reporting warnings when
the provider returned an unexpected value to try and help locate legacy
provider bugs, but any actual issues located from those warnings were
very few and far between.
Because the prior state cannot be reliably decoded when faced with
incompatible provider schema upgrades, and there is no longer any
significant reason to try and get the prior state at all, we can skip
the process entirely.
Data sources do not have state migrations, so there may be no way to
decode the prior state when faced with incompatible type changes.
Because prior state is only informational to the plan, and its existence
should not effect the planning process, we can skip decoding when faced
with errors.
When rendering diffs for resources which use nested attribute types, we
must cope with collections backing those attributes which are entirely
sensitive. The most common way this will be seen is through sensitive
values being present in sets, which will result in the entire set being
marked sensitive.
