| Age | Commit message (Collapse) | Author |
|---|
|
As specified in the Go specification, adapted to the following:
- the scanner recognise blocks as tokens
- the scanner is multi-language: define keywords in scanner spec
- as a result, we define how to skip semi-colon insertion rather
than how to add it.
|
|
The scanner returns a slice of pointers to tokens instead of a
slice of tokens. The parser now pass the initial node context.
|
|
Refctor node kind names by concatenating category and instance, to
allow better sorting. Comments are now parsed and skipped during
generation of AST.
|
|
So multiple successive incremental Evals function correctly.
Also improve the following:
- Apply the same Eval API to vm0 and vm1
- parser: dot diagram display is now synchronous
- codegen: outsource complex code generation for readability
- vm1: Pop take the number of values to pop as operand
|
|
|
|
|
|
|
|
* codegen: add a bytecode generator
* cleaning scanner, parser and vm1.
|
|
* parser: define all node kinds to make the parser multi-language
Defining all AST node kinds in the parser is necessary to make the
parser really multi-language. If a language requires a node kind not
already present in parser/kind.go, it will be necessary to add it first
here.
Note that as long as a node kind subtree is structurally identical
between languages, even if there are lexical and/or syntaxic
differences, it can (and must) be shared amongst multiple language
definitions.
For example, an "if" statememt in shell script or in C code should give
the same `IfStmt` at AST level.
In order to let the parser deal with the various language syntaxes,
and produce the right node kind and subtree, parser flags will be set
in language definitions (see `Flags` field in `NodeSpec` struct).
* lang/golang: use parser node kinds
* vm0: remode dependency on language definition.
|
|
|
