- feat: return an ExprSyntaxError for invalid references that end in a dot (#692)
- Introduce
ParseTraversalPartial, which allows traversals that include the splat ([*]) index operator. (#673) - ext/dynblock: Now accepts marked values in
for_each, and will transfer those marks (as much as technically possible) to values in the generated blocks. (#679)
- Expression evaluation will no longer panic if the splat operator is applied to an unknown value that has cty marks. (#678)
- Return
ExprSyntaxErrorwhen an invalid namespaced function is encountered during parsing (#668)
- Standardize on only two value dumping/diffing libraries (#669)
- Support for namespaced functions (#639)
- ext/dynblock: if
iteratoris invalid return this error instead of consequential errors (#656)
- ext/dynblock:
dynblock.Expandnow supports an optional hook for calling applications to check and potentially veto (by returning error diagnostics) particularfor_eachvalues. The behavior is unchanged for callers that don't set the new option. (#634)
- hclsyntax: Further fixes for treatment of "marked" values in the conditional expression, and better tracking of refined values into the conditional expression results, building on the fixes from v2.18.1. (#633)
- hclsyntax: Conditional expressions will no longer panic when one or both of their results are "marked", as is the case for situations like how HashiCorp Terraform tracks its concept of "sensitive values". (#630)
-
HCL now uses the tables from Unicode 15 when performing string normalization and character segmentation. HCL was previously using the Unicode 13 tables.
For calling applications where consistent Unicode support is important, consider also upgrading to Go 1.21 at the same time as adopting HCL v2.18.0 so that the standard library unicode tables (used for case folding, etc) will also be from Unicode 15.
- hclsyntax: When evaluating string templates that have a long known constant prefix, HCL will truncate the known prefix to avoid creating excessively-large refinements. String prefix refinements are intended primarily for relatively-short fixed prefixes, such as
https://at the start of a URL known to use that scheme. (#617) - ext/tryfunc: The "try" and "can" functions now handle unknown values slightly more precisely, and so can return known values in more situations when given expressions referring to unknown symbols. (#622)
- ext/typeexpr: Will no longer try to refine unknown values of unknown type when dealing with a user-specified type constraint containing the
anykeyword, avoiding an incorrect panic at runtime. (#625) - ext/typeexpr: Now correctly handles attempts to declare the same object type attribute multiple times by returning an error. Previously this could potentially panic by creating an incoherent internal state. (#624)
-
HCL now uses a newer version of the upstream
ctylibrary which has improved treatment of unknown values: it can now track additional optional information that reduces the range of an unknown value, which allows some operations against unknown values to return known or partially-known results. (#590)Note: This change effectively passes on
cty's notion of backward compatibility whereby unknown values can become "more known" in later releases. In particular, if your caller is usingcty.Value.RawEqualsin its tests against the results of operations with unknown values then you may see those tests begin failing after upgrading, due to the values now being more "refined".If so, you should review the refinements with consideration to the
ctyrefinements docs and update your expected results to match only if the reported refinements seem correct for the given situation. TheRawEqualsmethod is intended only for making exact value comparisons in test cases, so main application code should not use it; useEqualsinstead for real logic, which will take refinements into account automatically.
- ext/typeexpr: Verify type assumptions when applying default values, and ignore input values that do not match type assumptions. (#594)
- hclsyntax: Report correct
Range.EndforFunctionCallwith incomplete argument (#588)
-
ext/typeexpr: Modify the
Defaultsfunctionality to implement additional flexibility. HCL will now upcast lists and sets into tuples, and maps into objects, when applying default values if the applied defaults cause the elements within a target collection to have differing types. Previously, this would have resulted in a panic, now HCL will return a modified overall type. (#574)Users should return to the advice provided by v2.14.0, and apply the go-cty convert functionality after setting defaults on a given
cty.Value, rather than before. -
hclfmt: Avoid rewriting unchanged files. (#576)
-
hclsyntax: Simplify the AST for certain string expressions. (#584)
- hclwrite: Fix data race in
formatSpaces. (#511)
- ext/typeexpr: Skip null objects when applying defaults. This prevents crashes when null objects are creating inside collections, and stops incomplete objects being created with only optional attributes set. (#567)
- ext/typeexpr: Ensure default values do not have optional metadata attached. This prevents crashes when default values are inserted into concrete go-cty values that have also been stripped of their optional metadata. (#568)
- ext/typeexpr: With the go-cty upstream depenendency updated to v1.12.0, the
Defaultsstruct and associated functions can apply additional and more flexible 'unsafe' conversions (examples include tuples into collections such as lists and sets, and additional safety around null and dynamic values). (#564) - ext/typeexpr: With the go-cty upstream depenendency updated to v1.12.0, users should now apply the go-cty convert functionality before setting defaults on a given
cty.Value, rather than after, if they require a specificcty.Type. (#564)
- ext/typeexpr: Type convert defaults for optional object attributes when applying them. This prevents crashes in certain cases when the objects in question are part of a collection. (#555)
- ext/typeexpr: Added support for optional object attributes to
TypeConstraint. Attributes can be wrapped in the specialoptional(…)modifier, allowing the attribute to be omitted while still meeting the type constraint. For more information, cty's documentation on conversion between object types. (#549) - ext/typeexpr: New function:
TypeConstraintWithDefaults. In this mode, theoptional(…)modifier accepts a second argument which can be used as the default value for omitted object attributes. The function returns both acty.Typeand associatedDefaults, the latter of which has anApplymethod to apply defaults to a given value. (#549)
- hcl:
hcl.Diagnosticnow has an additional fieldExtrawhich is intended for carrying arbitrary supporting data ("extra information") related to the diagnostic message, intended to allow diagnostic renderers to optionally tailor the presentation of messages for particular situations. (#539) - hclsyntax: When an error occurs during a function call, the returned diagnostics will include extra information (as described in the previous point) about which function was being called and, if the message is about an error returned by the function itself, that raw
errorvalue without any post-processing. (#539)
- hclwrite: Fixed a potential data race for any situation where
hclwrite.Formatruns concurrently with itself. (#534)
- hclsyntax: Evaluation of conditional expressions will now produce more precise error messages about inconsistencies between the types of the true and false result expressions, particularly in cases where both are of the same structural type kind but differ in their nested elements. (#530)
- hclsyntax: The lexer will no longer allocate a small object on the heap for each token. Instead, in that situation it will allocate only when needed to return a diagnostic message with source location information. (#490)
- hclwrite: New functions
TokensForTuple,TokensForObject, andTokensForFunctionCallallow for more easily constructing the three constructs which are supported for static analysis and which HCL-based languages typically use in contexts where an expression is used only for its syntax, and not evaluated to produce a real value. For example, these new functions together are sufficient to construct all valid type constraint expressions from the Type Expressions Extension, which is the basis of variable type constraints in the Terraform language at the time of writing. (#502) - json: New functions
IsJSONExpressionandIsJSONBodyto determine if a given expression or body was created by the JSON syntax parser. In normal situations it's better not to worry about what syntax a particular expression/body originated in, but this can be useful in some trickier cases where an application needs to shim for backwards-compatibility or for static analysis that needs to have special handling of the JSON syntax's embedded expression/template conventions. (#524)
- gohcl: Fix docs about supported types for blocks. (#507)
- hclsyntax: The type for an upgraded unknown value with a splat expression cannot be known (#495)
- hclsyntax: Various error messages related to unexpectedly reaching end of file while parsing a delimited subtree will now return specialized messages describing the opening tokens as "unclosed", instead of returning a generic diagnostic that just happens to refer to the empty source range at the end of the file. This gives better feedback when error messages are being presented alongside a source code snippet, as is common in HCL-based applications, because it shows which innermost container the parser was working on when it encountered the error. (#492)
- hclsyntax: Upgrading an unknown single value to a list using a splat expression must return unknown (#493)
- dynblock: Decode unknown dynamic blocks in order to obtain any diagnostics even though the decoded value is not used (#476)
- hclsyntax: Calling functions is now more robust in the face of an incorrectly-implemented function which returns a
function.ArgErrorwhose argument index is out of range for the length of the arguments. Previously this would often lead to a panic, but now it'll return a less-precice error message instead. Functions that return out-of-bounds argument indices still ought to be fixed so that the resulting error diagnostics can be as precise as possible. (#472) - hclsyntax: Ensure marks on unknown values are maintained when processing string templates. (#478)
- hcl: Improved error messages for various common error situtions in
hcl.Indexandhcl.GetAttr. These are part of the implementation of indexing and attribute lookup in the native syntax expression language too, so the new error messages will apply to problems using those operators. (#474)
- dynblock,hcldec: Using dynblock in conjunction with hcldec can now decode blocks with unknown dynamic for_each arguments as entirely unknown values (#461)
- hclsyntax: Some syntax errors during parsing of the inside of
${...}template interpolation sequences will now produce an extra hint message about the need to escape as$${when trying to include interpolation syntax for other languages like shell scripting, AWS IAM policies, etc. (#462)
- hclsyntax: Fix panic for marked index value. (#451)
- HCL's native syntax and JSON scanners -- and thus all of the other parsing components that build on top of them -- are now using Unicode 13 rules for text segmentation when counting text characters for the purpose of reporting source location columns. Previously HCL was using Unicode 12. Unicode 13 still uses the same algorithm but includes some additions to the character tables the algorithm is defined in terms of, to properly categorize new characters defined in Unicode 13.
- hclsyntax: Fix panic for marked collection splat. (#436)
- hclsyntax: Fix panic for marked template loops. (#437)
- hclsyntax: Fix
forexpression marked conditional. (#438) - hclsyntax: Mark objects with keys that are sensitive. (#440)
- hclsyntax: Fix panic when expanding marked function arguments. (#429)
- hclsyntax: Error when attempting to use a marked value as an object key. (#434)
- hclsyntax: Error when attempting to use a marked value as an object key in expressions. (#433)
- hclsyntax: Expression grouping parentheses will now be reflected by an explicit node in the AST, whereas before they were only considered during parsing. (#426)
- hclwrite: The parser will now correctly include the
(and)tokens when an expression is surrounded by parentheses. Previously it would incorrectly recognize those tokens as being extraneous tokens outside of the expression. (#426) - hclwrite: The formatter will now remove (rather than insert) spaces between the
!(unary boolean "not") operator and its subsequent operand. (#403) - hclsyntax: Unmark conditional values in expressions before checking their truthfulness (#427)
- gohcl: Fix panic when decoding into type containing value slices. (#335)
- hclsyntax: The unusual expression
null[*]was previously always returning an unknown value, even though the rules for[*]normally call for it to return an empty tuple when applied to a null. As well as being a surprising result, it was particularly problematic because it violated the rule that a calling application may assume that an expression result will always be known unless the application itself introduces unknown values via the evaluation context.null[*]will now produce an empty tuple. (#416) - hclsyntax: Fix panic when traversing a list, tuple, or map with cty "marks" (#424)
- hclwrite: Correctly handle blank quoted string block labels, instead of dropping them (#422)
- json: There is a new function
ParseWithStartPos, which allows overriding the starting position for parsing in case the given JSON bytes are a fragment of a larger document, such as might happen when decoding withencoding/jsoninto ajson.RawMessage. (#389) - json: There is a new function
ParseExpression, which allows parsing a JSON string directly in expression mode, whereas previously it was only possible to parse a JSON string in body mode. (#381) - hclwrite:
Blocktype now supportsSetTypeandSetLabels, allowing surgical changes to the type and labels of an existing block without having to reconstruct the entire block. (#340)
- hclsyntax: Fix confusing error message for bitwise OR operator (#380)
- hclsyntax: Several bug fixes for using HCL with values containing cty "marks" (#404, #406, #407)
- hcldec: Add a new
Spec,ValidateSpec, which allows custom validation of values at decode-time. (#387)
- hclsyntax: Fix panic with combination of sequences and null arguments (#386)
- hclsyntax: Fix handling of unknown values and sequences (#386)
- hclwrite: handle legacy dot access of numeric indexes. (#369)
- hclwrite: Fix panic for dotted full splat (
foo.*) (#374)
- hclwrite: Generate multi-line objects and maps. (#372)
- The Unicode data tables that HCL uses to produce user-perceived "column" positions in diagnostics and other source ranges are now updated to Unicode 12.0.0, which will cause HCL to produce more accurate column numbers for combining characters introduced to Unicode since Unicode 9.0.0.
- json: Fix panic when parsing malformed JSON. (#358)
- ext/tryfunc: Optional functions
tryandcanto include in yourhcl.EvalContextwhen evaluating expressions, which allow users to make decisions based on the success of expressions. (#330) - ext/typeexpr: Now has an optional function
convertwhich you can include in yourhcl.EvalContextwhen evaluating expressions, allowing users to convert values to specific type constraints using the type constraint expression syntax. (#330) - ext/typeexpr: A new
ctycapsule typetypeexpr.TypeConstraintTypewhich, when used as either a type constraint for a function parameter or as a type constraint for ahcldecattribute specification will cause the given expression to be interpreted as a type constraint expression rather than a value expression. (#330) - ext/customdecode: An optional extension that allows overriding the static decoding behavior for expressions either in function arguments or
hcldecattribute specifications. (#330) - ext/customdecode: New
ctycapsuletypescustomdecode.ExpressionTypeandcustomdecode.ExpressionClosureTypewhich, when used as either a type constraint for a function parameter or as a type constraint for ahcldecattribute specification will cause the given expression (and, for the closure type, also thehcl.EvalContextit was evaluated in) to be captured for later analysis, rather than immediately evaluated. (#330)
- hcldec: Attribute evaluation (as part of
AttrSpecorBlockAttrsSpec) now captures expression evaluation metadata in any errors it produces during type conversions, allowing for better feedback in calling applications that are able to make use of this metadata when printing diagnostic messages. (#329)
- hclsyntax:
IndexExpr,SplatExpr, andRelativeTraversalExprwill now report a source range that covers all of their child expression nodes. Previously they would report only the operator part, such as["foo"],[*], or.foo, which was problematic for callers using source ranges for code analysis. (#328) - hclwrite: Parser will no longer panic when the input includes index, splat, or relative traversal syntax. (#328)
- gohcl: When decoding into a struct value with some fields already populated, those values will be retained if not explicitly overwritten in the given HCL body, with similar overriding/merging behavior as
json.Unmarshalin the Go standard library. - hclwrite: New interface to set the expression for an attribute to be a raw token sequence, with no special processing. This has some caveats, so if you intend to use it please refer to the godoc comments. (#320)
- hclwrite: The
Body.Blocksmethod was returing the blocks in an indefined order, rather than preserving the order of declaration in the source input. (#313) - hclwrite: The
TokensForTraversalfunction (and thus in turn theBody.SetAttributeTraversalmethod) was not correctly handling index steps in traversals, and thus producing invalid results. (#319)
Initial release of HCL 2, which is a new implementating combining the HCL 1 language with the HIL expression language to produce a single language supporting both nested configuration structures and arbitrary expressions.
HCL 2 has an entirely new Go library API and so is not a drop-in upgrade relative to HCL 1. It's possible to import both versions of HCL into a single program using Go's semantic import versioning mechanism:
import (
hcl1 "github.com/hashicorp/hcl"
hcl2 "github.com/hashicorp/hcl/v2"
)
Prior to v2.0.0 there was not a curated changelog. Consult the git history from the latest v1.x.x tag for information on the changes to HCL 1.