This exercises the ability to customize the installation methods used by
the provider plugin installer, in this case forcing the use of a custom
local directory with a result essentially the same as what happens when
you pass -plugin-dir to "terraform init".
The CLI config can be written in both native HCL and HCL JSON syntaxes, so
the provider_installation block must be expressible using JSON too. Our
previous checks to approximate HCL 2-level strictness were too strict for
HCL JSON where things are more ambiguous even in HCL 2, so this includes
some additional relaxations if we detect that we're decoding an AST
produced from a JSON file.
This is still subject to the quirky ways HCL 1 handles JSON though, so
the JSON value must be structured in a way that doesn't trigger HCL's
heuristics that try to guess what is a block and what is an attribute.
(This is the issue that HCL 2 fixes by always decoding using a schema;
there's more context on this in:
https://log.martinatkins.me/2019/04/25/hcl-json/ )
Unfortunately in the user model the noun "source" is already used for the
argument in the required_providers block to specify which provider to use,
so it's confusing to use the same noun to also refer to the method used to
obtain that provider.
In the hope of mitigating that confusion, here we use the noun "method",
as in "installation method", to talk about the decision between getting
a provider directly from its origin registry or getting it from some
mirror. This is distinct from the provider's "source", which is the
location where a provider _originates_ (prior to mirroring).
This noun is also not super awesome, but better than overloading an
existing term in the same feature.
In the first pass of implementing this it was strict about what arguments
are allowed inside source blocks, but that was counter to our usual design
principles for CLI config where we tend to ignore unrecognized things to
allow for some limited kinds of future expansion without breaking
compatibility with older versions of Terraform that will be sharing the
same CLI configuration files with newer versions.
However, I'd removed the tracking of that prior to the initial commit. I
missed some leftover parts when doing that removal, so this cleans up the
rest of it.
An earlier commit added a redundant stub for a new network mirror source
that was already previously stubbed as HTTPMirrorSource.
This commit removes the unnecessary extra stub and changes the CLI config
handling to use it instead. Along the way this also switches to using a
full base URL rather than just a hostname for the mirror, because using
the usual "Terraform-native service discovery" protocol here doesn't isn't
as useful as in the places we normally use it (the mirror mechanism is
already serving as an indirection over the registry protocol) and using
a direct base URL will make it easier to deploy an HTTP mirror under
a path prefix on an existing static file server.
When we originally introduced this environment variable it was intended to
solve for the use-case where a particular invocation of Terraform needs
a different CLI configuration than usual, such as if Terraform is being
run as part of an automated test suite or other sort of automated
situation with different needs than normal use.
However, we accidentally had it only override the original singleton CLI
config file, while leaving the CLI configuration directory still enabled.
Now we'll take the CLI configuration out of the equation too, so that only
the single specified configuration file and any other environment-sourced
settings will be included.
If the CLI configuration contains a provider_installation block then we'll
use the source configuration it describes instead of the implied one we'd
build otherwise.
This is a placeholder for later implementation of a mirror source that
talks to a particular remote HTTP server and expects it to implement the
provider mirror protocol.
This new CLI config block type allows explicitly specifying where
Terraform should look to find provider plugins for installation. This is
not used anywhere as of this commit, but in a future commit we'll change
package main to treat the presence of a block of this type as a request
to disable the default set of provider sources and use these explicitly-
specified ones instead.
A side effect of the various changes to the provider installer included losing the initialization required error message which would occur if a user removed or modified the .terraform directory.
Previously, plugin factories were created after the configuration was loaded, in terraform.NewContext. Terraform would compare the required providers (from config and state) to the available providers and return the aforementioned error if a provider was missing.
Provider factories are now loaded at the beginning of any terraform command, before terraform even loads the configuration, and therefore before terraform has a list of required providers.
This commit replaces the current error when a providers' schema cannot be found in the provider factories with the init error, and adds a command test (to plan tests, for no real reason other than that's what I thought of first).
This more closely replicates the 0.12-and-earlier behavior, where having
at least one version of a provider installed locally would totally disable
any attempt to look for newer versions remotely.
This is just for the implicit default behavior. Assumption is that later
we'll have an explicit configuration mechanism that will allow the user
to specify exactly where to look for what, and thus avoid tricky
heuristics like this.
Providers installed from the registry are accompanied by a list of
checksums (the "SHA256SUMS" file), which is cryptographically signed to
allow package authentication. The process of verifying this has multiple
steps:
- First we must verify that the SHA256 hash of the package archive
matches the expected hash. This could be done for local installations
too, in the future.
- Next we ensure that the expected hash returned as part of the registry
API response matches an entry in the checksum list.
- Finally we verify the cryptographic signature of the checksum list,
using the public keys provided by the registry.
Each of these steps is implemented as a separate PackageAuthentication
type. The local archive installation mechanism uses only the archive
checksum authenticator, and the HTTP installation uses all three in the
order given.
The package authentication system now also returns a result value, which
is used by command/init to display the result of the authentication
process.
There are three tiers of signature, each of which is presented
differently to the user:
- Signatures from the embedded HashiCorp public key indicate that the
provider is officially supported by HashiCorp;
- If the signing key is not from HashiCorp, it may have an associated
trust signature, which indicates that the provider is from one of
HashiCorp's trusted partners;
- Otherwise, if the signature is valid, this is a community provider.
The providers command has been refactored to use the modern provider types and
ProviderRequirements() functions. This resulted in a breaking change to
the output: it no longer outputs the providers by module and no longer
prints `(inherited)` or `(from state)` to show why a provider is
included. We decided that at this time it was best to stick with the
existing functions and make this change, but if we get feedback from the
community we will revisit.
Additional tests to exercise providers in modules and providers from
state have been included.
This PR adds iteration through any provider configuration blocks in the
config in addProviderRequirements().
A stale comment (of mine!) would leave one expecting the
module.ProviderRequirements to include any requirements from provider
configs. The comment was inaccurate and has been updated.
Martin Atkins