cty now guarantees that sets of primitive values will iterate in a
reasonable order. Previously it was the caller's responsibility to deal
with that, but we invariably neglected to do so, causing inconsistent
ordering. Since cty prioritizes consistent behavior over performance, it
now imposes its own sort on set elements as part of iterating over them so
that calling applications don't have to worry so much about it.
This change also causes cty to consistently push unknown and null values
in sets to the end of iteration, where before that was undefined. This
means that our diff output will now consistently list additions before
removals when showing sets, rather than the ordering being undefined as
before.
The ordering of known, non-null, non-primitive values is still not
contractually fixed but remains consistent for a particular version of
cty.
By upgrading this to a version that supports Go Modules, we can avoid
the redundant dependencies labix.org/v2/mgo and launchpad.net/gocheck.
These dependencies come via hashicorp/consul, which we have vendored for
the Consul backend.
There are no changes to the go-msgpack code here; we're just adopting a
newer version that has a go.mod present so that its dependencies can be
determined more accurately.
This gives us an extra hook in the dynblock variables analysis that should
allow us to also make it subject also to the lang/blocktoattr fixup, to
ensure we'll find all the references in spite of these various
pre-processing wrappers.
This includes a fix to prevent unintentional infinite recursion when
trying to unify multiple object types to a single type for conversion to
list(any).
Sadly I wasn't able to reproduce the problem as reported (in #20728), so
therefore I wasn't able to write a Terraform test for it, but I have
confirmed that the cty behavior here was incorrect anyway (recursively
calling the same function we're already in with the same arguments is
clearly not productive) and so this change will allow whatever situation
that was to terminate with a type conversion error, rather than causing a
stack overflow.
It's likely that there is another bug lurking under this, since the
problematic code here was supposed to be unreachable, but avoiding the
crash is the priority for now. If the problem re-surfaces then it should
at least be an error message with some additional context about what the
goal of the caller was.
This also includes an unrelated fix for the gocty package, which doesn't
affect Terraform because it makes very little use of that package.
This includes improved functionality for HCL's "dynamic block extension",
which will allow us (in a subsequent commit) to properly detect
dependencies inside nested "dynamic" blocks, where currently they get
missed.
For this commit though, we just upgrade HCL to a version that includes it
and make a small change to our "lang" package to align with an upstream
renaming.
This includes two upstream fixes:
- Handle explicit JSON "null" consistently during decode of JSON syntax.
- Properly detect the end of a "heredoc" when formatting to avoid messing
up indentation of other lines following the heredoc.
This includes a fix for the parsing of object for expressions in newline-
sensitive contexts like block bodies.
It also includes a change to the JSON syntax decoder that cause it to
consider an explicit null to be equivalent to a property not being set at
all when interpreting a property value as a nested block. (It was
previously doing tha only when interpreting the property value as an
attribute value.)
Martin Atkins