Although this end symbol appears as just a close-brace in source, it's
worth differentiating it because the scanner must differentiate it anyway
(to recognize moving back into template-scanning mode) and it avoids the
parser from having to similarly re-recognize the difference.
On reflection, it seems easier to maintain the necessary state we need
by doing all of the scanning in a single pass, since we can then just
use local variables within the scanner function.
Using Ragel here because the scanner is going to be somewhat complex due
to the need to switch back and forth between normal and template states,
etc. This should be easier to maintain than a hand-written scanner, while
ragel gives us the extra features we need to implement things that would
normally be too complex for a "regular" scanner generator.
This means we can actually point at a column in the console without it
getting misaligned by multi-byte UTF-8 sequences and Unicode combining
characters.
This is the first non-trivial expression Value implementation. Lots of
code here, so hopefully while implementing other expressions some
opportunities emerge to factor out some of these details.
The implementation of Variables will be identical for every Expression
implementation since we just wrap our AST-walk-based "Variables" function
to do the work.
Rather than manually copy-pasting the declaration for each expression
type, instead we'll generate this programmatically using "go generate".
This will need to be re-run each time a new expression node type is
added, in order to make it actually implement the Expression interface.
This function is effectively the implementation of Variables for all
expressions, but unfortunately we still need to declare a wrapper around
it as a method on every single expression type.
This package will grow to contain all of the gory details of the native
zcl syntax, including it AST, parser, etc. Most callers should access
this via the simpler API in the top-level package, which then gives
automatic support for other syntaxes too.
Traversals are a generalized way to talk about paths taken from the scope
and from arbitrary values. These will be used for various analysis tasks,
such as determining what needs to be placed into a scope.
Eventually zcl will have its own native template format that we'll use
by default, but for applications migrating from HCL/HIL we can instead
parse strings as HIL templates, for compatibility with how JSON configs
would've been processed in a HCL/HIL app.
When this mode is not enabled, we still just treat strings as literals,
pending the implementation of the zcl template parser.
Currently only deals with the literal HCL structure. Later we will also
support HIL parsing and evaluation within strings, to achieve parity with
existing uses of HCL/HIL together. For now, this has parity with uses of
HCL alone, with the exception that float and int values are not
distinguished, because cty does not make this distinction.
This is intended as a backward-compatibility interface, allowing
applications that previously used HCL/HIL to adopt zcl while still being
able to parse their old HCL/HIL-based configuration file formats.
When producing diagnostics about missing attributes or bodies it's
necessary to have a range representing the place where the missing thing
might be inserted.
There's not always a single reasonable value for this, so some liberty
can be taken about what exactly is returned as long as it's somewhere
the user can relate back to the construct producing the error.
This is really just a wrapper around converting the value to a Go native
value with gocty, but is convenient to provide since it can attempt
type conversion and produce diagnostics automatically.
Expressions can now be evaluated within an "EvalContext", which provides
the variable and function scopes. The JSON implementation of this
currently ignores the context entirely and just returns literal values,
since we've not yet implemented the template language parser that would
be needed for the JSON parser to properly support expressions.
The Content and PartialContent methods deal with the case where the caller
knows what structure is expected within the body, but sometimes the
structure of a body is just a free-form set of attributes that the caller
needs to enumerate.
The idea here is that the block in question must contain only attributes,
and no child blocks. For JSON this just entails interpreting every
property as an attribute. For native syntax later this will mean
producing an error diagnostic if any blocks appear within the body.
When using the parser to do static analysis and editor integrations, it's
still useful to be able to get the incomplete AST resulting from a parse
error.
Callers that intend to use the returned value to take real actions (as
opposed to just analysis) must check diags.HasError() to determine if
the returned file can be considered valid.
Its implementation calls into each of the child bodies in turn and merges
the result to produce a single BodyContent. This is intended to support
the case of a directory being the unit of configuration rather than a
file, with the calling application discovering and parsing each of the
files in its workspace and then merging them together for processing as
a single configuration.
We need to be careful to keep straight the distinction between JSON
properties and zcl body attributes here, since properties can represent
both attributes _and_ blocks.
This is a wrapper around Body.PartialContent that generates additional
error diagnostics if any object properties are left over after decoding,
helping the config author to catch typos that would otherwise have caused
a property to be silently ignored.
