Here we propagate in the initialized terminal.Streams from package main,
and then onwards to backends running in CLI mode.
This also replaces our use of helper/wrappedstreams to determine whether
stdin is a terminal or a pipe. helper/wrappedstreams returns incorrect
file descriptors on Windows, causing StdinPiped to always return false on
that platform and thus causing one of the odd behaviors discussed in
Finally, this includes some wrappers around the ability to look up the
number of columns in the terminal in preparation for use elsewhere. These
wrappers deal with the fact that our unit tests typically won't populate
meta.Streams.
* command/format: concise diff is no longer an experiment
Since state formatting goes through the "diff" printer, I have
repurposed the concise flag as a verbose flag, used only when printing
state. It's silly but it works!
* remove helper/experiment
With this experiment concluded, we no longer need helper/experiment. The
shadow experiment had not been touched in many years, so I removed all
references, and removed the package entirely. Any new experiments are
expected to be configuration experiments handled by our (other)
experiments package.
* check for the verbose flag consistently, in case we end up using it in plans in the future
When using the enhanced remote backend, a subset of all Terraform
operations are supported. Of these, only plan and apply can be executed
on the remote infrastructure (e.g. Terraform Cloud). Other operations
run locally and use the remote backend for state storage.
This causes problems when the local version of Terraform does not match
the configured version from the remote workspace. If the two versions
are incompatible, an `import` or `state mv` operation can cause the
remote workspace to be unusable until a manual fix is applied.
To prevent this from happening accidentally, this commit introduces a
check that the local Terraform version and the configured remote
workspace Terraform version are compatible. This check is skipped for
commands which do not write state, and can also be disabled by the use
of a new command-line flag, `-ignore-remote-version`.
Terraform version compatibility is defined as:
- For all releases before 0.14.0, local must exactly equal remote, as
two different versions cannot share state;
- 0.14.0 to 1.0.x are compatible, as we will not change the state
version number until at least Terraform 1.1.0;
- Versions after 1.1.0 must have the same major and minor versions, as
we will not change the state version number in a patch release.
If the two versions are incompatible, a diagnostic is displayed,
advising that the error can be suppressed with `-ignore-remote-version`.
When this flag is used, the diagnostic is still displayed, but as a
warning instead of an error.
Commands which will not write state can assert this fact by calling the
helper `meta.ignoreRemoteBackendVersionConflict`, which will disable the
checks. Those which can write state should instead call the helper
`meta.remoteBackendVersionCheck`, which will return diagnostics for
display.
In addition to these explicit paths for managing the version check, we
have an implicit check in the remote backend's state manager
initialization method. Both of the above helpers will disable this
check. This fallback is in place to ensure that future code paths which
access state cannot accidentally skip the remote version check.
When building a context, we read the dependency locks and ensure that
the provider requirements from the configuration can be satisfied.
If the configured requirements change such that the locks need to be
updated, we explain this and recommend running "terraform init".
This check is ignored for any providers which are locally marked as in
development. This includes unmanaged providers and those listed in the
provider installation `dev_overrides` block.
For normal provider installation we want to associate each provider with
a selected version number and find a suitable package for that version
that conforms to the official hashes for that release.
Those requirements are very onerous for a provider developer currently
testing a not-yet-released build, though. To allow for that case this new
CLI configuration feature allows overriding specific providers to refer
to give local filesystem directories.
Any provider overridden in this way is not subject to the usual
restrictions about selected versions or checksum conformance, and
activating an override won't cause any changes to the selections recorded
in the lock file because it's intended to be a temporary setting for one
developer only.
This is, in a sense, a spiritual successor of an old capability we had to
override specific plugins in the CLI configuration file. There were
some vestiges of that left in the main package and CLI config package
but nothing has actually been honoring them for several versions now and
so this commit removes them to avoid confusion with the new mechanism.
This changes the approach used by the provider installer to remember
between runs which selections it has previously made, using the lock file
format implemented in internal/depsfile.
This means that version constraints in the configuration are considered
only for providers we've not seen before or when -upgrade mode is active.
In earlier commits we started to make the installation codepath
context-aware so that it could be canceled in the event of a SIGINT, but
we didn't complete wiring that through the API of the getproviders
package.
Here we make the getproviders.Source interface methods, along with some
other functions that can make network requests, take a context.Context
argument and act appropriately if that context is cancelled.
The main providercache.Installer.EnsureProviderVersions method now also
has some context-awareness so that it can abort its work early if its
context reports any sort of error. That avoids waiting for the process
to wind through all of the remaining iterations of the various loops,
logging each request failure separately, and instead returns just
a single aggregate "canceled" error.
We can then set things up in the "terraform init" and
"terraform providers mirror" commands so that the context will be
cancelled if we get an interrupt signal, allowing provider installation
to abort early while still atomically completing any local-side effects
that may have started.
This new option is intended to address the previous inconsistencies where
some older subcommands supported partially changing the target directory
(where Terraform would use the new directory inconsistently) where newer
commands did not support that override at all.
Instead, now Terraform will accept a -chdir command at the start of the
command line (before the subcommand) and will interpret it as a request
to direct all actions that would normally be taken in the current working
directory into the target directory instead. This is similar to options
offered by some other similar tools, such as the -C option in "make".
The new option is only accepted at the start of the command line (before
the subcommand) as a way to reflect that it is a global command (not
specific to a particular subcommand) and that it takes effect _before_
executing the subcommand. This also means it'll be forced to appear before
any other command-specific arguments that take file paths, which hopefully
communicates that those other arguments are interpreted relative to the
overridden path.
As a measure of pragmatism for existing uses, the path.cwd object in
the Terraform language will continue to return the _original_ working
directory (ignoring -chdir), in case that is important in some exceptional
workflows. The path.root object gives the root module directory, which
will always match the overriden working directory unless the user
simultaneously uses one of the legacy directory override arguments, which
is not a pattern we intend to support in the long run.
As a first step down the deprecation path, this commit adjusts the
documentation to de-emphasize the inconsistent old command line arguments,
including specific guidance on what to use instead for the main three
workflow commands, but all of those options remain supported in the same
way as they were before. In a later commit we'll make those arguments
produce a visible deprecation warning in Terraform's output, and then
in an even later commit we'll remove them entirely so that -chdir is the
single supported way to run Terraform from a directory other than the
one containing the root module configuration.
The workspace name can be overridden by setting a TF_WORKSPACE
environment variable. If this is done, we should still validate the
resulting workspace name; otherwise, we could end up with an invalid and
unselectable workspace.
This change updates the Meta.Workspace function to return an error, and
handles that error wherever necessary.
This adds supports for "unmanaged" providers, or providers with process
lifecycles not controlled by Terraform. These providers are assumed to
be started before Terraform is launched, and are assumed to shut
themselves down after Terraform has finished running.
To do this, we must update the go-plugin dependency to v1.3.0, which
added support for the "test mode" plugin serving that powers all this.
As a side-effect of not needing to manage the process lifecycle anymore,
Terraform also no longer needs to worry about the provider's binary, as
it won't be used for anything anymore. Because of this, we can disable
the init behavior that concerns itself with downloading that provider's
binary, checking its version, and otherwise managing the binary.
This is all managed on a per-provider basis, so managed providers that
Terraform downloads, starts, and stops can be used in the same commands
as unmanaged providers. The TF_REATTACH_PROVIDERS environment variable
is added, and is a JSON encoding of the provider's address to the
information we need to connect to it.
This change enables two benefits: first, delve and other debuggers can
now be attached to provider server processes, and Terraform can connect.
This allows for attaching debuggers to provider processes, which before
was difficult to impossible. Second, it allows the SDK test framework to
host the provider in the same process as the test driver, while running
a production Terraform binary against the provider. This allows for Go's
built-in race detector and test coverage tooling to work as expected in
provider tests.
Unmanaged providers are expected to work in the exact same way as
managed providers, with one caveat: Terraform kills provider processes
and restarts them once per graph walk, meaning multiple times during
most Terraform CLI commands. As unmanaged providers can't be killed by
Terraform, and have no visibility into graph walks, unmanaged providers
are likely to have differences in how their global mutable state behaves
when compared to managed providers. Namely, unmanaged providers are
likely to retain global state when managed providers would have reset
it. Developers relying on global state should be aware of this.
Back when we first introduced provider versioning in Terraform 0.10, we
did the provider version resolution in terraform.NewContext because we
weren't sure yet how exactly our versioning model was going to play out
(whether different versions could be selected per provider configuration,
for example) and because we were building around the limitations of our
existing filesystem-based plugin discovery model.
However, the new installer codepath is new able to do all of the
selections up front during installation, so we don't need such a heavy
inversion of control abstraction to get this done: the command package can
select the exact provider versions and pass their factories directly
to terraform.NewContext as a simple static map.
The result of this commit is that CLI commands other than "init" are now
able to consume the local cache directory and selections produced by the
installation process in "terraform init", passing all of the selected
providers down to the terraform.NewContext function for use in
implementing the main operations.
This commit is just enough to get the providers passing into the
terraform.Context. There's still plenty more to do here, including to
repair all of the tests this change has additionally broken.
Following the same approach we use for other CLI-Config-able objects like
the service discovery system, the main package is responsible for
producing a suitable implementation of this interface which the command
package can then use.
When unit testing in the command package we can then substitute mocks as
necessary, following the dependency inversion principle.
When warnings appear in isolation (not accompanied by an error) it's
reasonable to want to defer resolving them for a while because they are
not actually blocking immediate work.
However, our warning messages tend to be long by default in order to
include all of the necessary context to understand the implications of
the warning, and that can make them overwhelming when combined with other
output.
As a compromise, this adds a new CLI option -compact-warnings which is
supported for all the main operation commands and which uses a more
compact format to print out warnings as long as they aren't also
accompanied by errors.
The default remains unchanged except that the threshold for consolidating
warning messages is reduced to one so that we'll now only show one of
each distinct warning summary.
Full warning messages are always shown if there's at least one error
included in the diagnostic set too, because in that case the warning
message could contain additional context to help understand the error.
Some of our warnings are produced in response to particular configuration
constructs which might appear many times across a Terraform configuration.
To avoid the warning output dwarfing all of the other output, we'll use
ConsolidateWarnings to limit each distinct warning summary to appear at
most twice, and annotate the final one in the sequence with an additional
paragraph noting that some number of them have been hidden.
This is intended as a compromise to ensure that these warnings are still
seen and noted but to help ensure that we won't produce so many of them
as to distract from other output that appears alongside them.
This applies only to warnings relating to specific configuration ranges;
errors will continue to be shown individually, and sourceless warnings
(which are rare in Terraform today) will likewise remain ungrouped because
they are less likely to be repeating the same statement about different
instances of the same problem throughout the configuration.
During the 0.12 work we intended to move all of the variable value
collection logic into the UI layer (command package and backend packages)
and present them all together as a unified data structure to Terraform
Core. However, we didn't quite succeed because the interactive prompts
for unset required variables were still being handled _after_ calling
into Terraform Core.
Here we complete that earlier work by moving the interactive prompts for
variables out into the UI layer too, thus allowing us to handle final
validation of the variables all together in one place and do so in the UI
layer where we have the most context still available about where all of
these values are coming from.
This allows us to fix a problem where previously disabling input with
-input=false on the command line could cause Terraform Core to receive an
incomplete set of variable values, and fail with a bad error message.
As a consequence of this refactoring, the scope of terraform.Context.Input
is now reduced to only gathering provider configuration arguments. Ideally
that too would move into the UI layer somehow in a future commit, but
that's a problem for another day.
For unit testing in particular we can't launch a real browser for testing,
so this indirection is primarily to allow us to substitute a mock when
testing a command that can launch a browser.
This includes a simple mock implementation that expects to interact with
a running web server directly.
Any command using meta.defaultFlagSet *might* occasionally exit before
the flag package's output got written. This caused flag error messages
to get lost. This PR discards the flag package output in favor of
directly returning the error to the end user.
Previously we were doing this rather inconsistently: some commands would
do it and others would not. By doing it here we ensure we always apply the
same normalization, regardless of which operation we're running.
This normalization is mostly for cosmetic purposes in error messages, but
it also ends up being used to populate path.module and path.root and so
it's important that we always produce consistent results here so that
we don't produce flappy changes as users work with different commands.
The fact that thus mutates a data structure as a side-effect is not ideal
but this is the best place to ensure it always gets applied without doing
any significant refactoring, since everything after this point happens in
the backend package where the normalizePath method is not available.
A lot of commands used `c.Meta.flagSet()` to create the initial flagset for the command, while quite a few of them didn’t actually use or support the flags that are then added.
So I updated a few commands to use `flag.NewFlagSet()` instead to only add the flags that are actually needed/supported.
Additionally this prevents a few commands from using locking while they actually don’t need locking (as locking is enabled as a default in `c.Meta.flagSet()`.
The mission of this process method used to include dealing with
auto-loaded tfvars files, but it doesn't do that anymore.
It does still deal with the -no-color option, but the test wasn't
exercising that part before.
Now the test here focuses on the -no-color behavior.
The process method still has a "vars" flag argument which is no longer
used. Since this is an unexported method we could potentially address this
but this commit is intentionally limited only to fixing the test.
This work was done against APIs that were already changed in the branch
before work began, and so it doesn't apply to the v0.12 development work.
To allow v0.12 to merge down to master, we'll revert this work out for now
and then re-introduce equivalent functionality in later commits that works
against the new APIs.
Due to how often the state and plan types are referenced throughout
Terraform, there isn't a great way to switch them out gradually. As a
consequence, this huge commit gets us from the old world to a _compilable_
new world, but still has a large number of known test failures due to
key functionality being stubbed out.
The stubs here are for anything that interacts with providers, since we
now need to do the follow-up work to similarly replace the old
terraform.ResourceProvider interface with its replacement in the new
"providers" package. That work, along with work to fix the remaining
failing tests, will follow in subsequent commits.
The aim here was to replace all references to terraform.State and its
downstream types with states.State, terraform.Plan with plans.Plan,
state.State with statemgr.State, and switch to the new implementations of
the state and plan file formats. However, due to the number of times those
types are used, this also ended up affecting numerous other parts of core
such as terraform.Hook, the backend.Backend interface, and most of the CLI
commands.
Just as with 5861dbf3fc49b19587a31816eb06f511ab861bb4 before, I apologize
in advance to the person who inevitably just found this huge commit while
spelunking through the commit history.
Because we gather together diagnostics from many different parts of the
codebase, the list often ends up being in a non-ideal order. Here we
define a partial ordering for diagnostics that should hopefully make them
easier to scan when many are present, by grouping together diagnostics
that are of the same severity and belong to the same file.
We use sort.Stable here because we have a partial order and so we need
to make sure that diagnostics that do not have a relative ordering will
remain in their original order.
This sorting is applied just in time before rendering the diagnostics
in command.Meta.showDiagnostics.
Due to how deeply the configuration types go into Terraform Core, there
isn't a great way to switch out to HCL2 gradually. As a consequence, this
huge commit gets us from the old state to a _compilable_ new state, but
does not yet attempt to fix any tests and has a number of known missing
parts and bugs. We will continue to iterate on this in forthcoming
commits, heading back towards passing tests and making Terraform
fully-functional again.
The three main goals here are:
- Use the configuration models from the "configs" package instead of the
older models in the "config" package, which is now deprecated and
preserved only to help us write our migration tool.
- Do expression inspection and evaluation using the functionality of the
new "lang" package, instead of the Interpolator type and related
functionality in the main "terraform" package.
- Represent addresses of various objects using types in the addrs package,
rather than hand-constructed strings. This is not critical to support
the above, but was a big help during the implementation of these other
points since it made it much more explicit what kind of address is
expected in each context.
Since our new packages are built to accommodate some future planned
features that are not yet implemented (e.g. the "for_each" argument on
resources, "count"/"for_each" on modules), and since there's still a fair
amount of functionality still using old-style APIs, there is a moderate
amount of shimming here to connect new assumptions with old, hopefully in
a way that makes it easier to find and eliminate these shims later.
I apologize in advance to the person who inevitably just found this huge
commit while spelunking through the commit history.
This is a rather-messy, complex change to get the "command" package
building again against the new backend API that was updated for
the new configuration loader.
A lot of this is mechanical rewriting to the new API, but
meta_config.go and meta_backend.go in particular saw some major
changes to interface with the new loader APIs and to deal with
the change in order of steps in the backend API.
We need to share a single config loader across all callers because that
allows us to maintain the source code cache we'll use for snippets in
error messages.
Nothing calls this yet. Callers will be gradually updated away from Module
and Config in subsequent commits.
If we get a diagnostic message that references a source range, and if the
source code for the referenced file is available, we'll show a snippet of
the source code with the source range highlighted.
At the moment we have no cache of source code, so in practice this
codepath can never be visited. Callers to format.Diagnostic will be
gradually updated in subsequent commits.
By adding this method you now only have to pass a `*disco.Disco` object around in order to do discovery and use any configured credentials for the discovered hosts.
Of course you can also still pass around both a `*disco.Disco` and a `auth.CredentialsSource` object if there is a need or a reason for that!
There's no point in trying to track these, they're lost after each test.
Kill them after a short delay so we don't have goroutines from every single
command test to wade through if we have a stack dump.
Only check for input twice in the meta.confirm method. This prevents an
errant newline from aborting the run while allowing Terraform to exit if
there is no input available. We don't just check for a tty, since we
still rely on being able to pipe input in for testing.
Remove the redundant confirmation loops in the migration code, and only
use the confirm method.
Now that the local backend can be cancelled during plan and refresh, we
don't really need the testShutdownHook. Simplify the tests by just
checking for Stop being called on the provider.
