@backbencher/brain 0.0.0-stage → 0.1.1

package/grammars/tree-sitter-vba/README.md

@@ -0,0 +1,738 @@
+# tree-sitter-vba
+
+[![npm version](https://img.shields.io/npm/v/tree-sitter-vba.svg)](https://www.npmjs.com/package/tree-sitter-vba)
+[![npm downloads](https://img.shields.io/npm/dm/tree-sitter-vba.svg)](https://www.npmjs.com/package/tree-sitter-vba)
+[![CI](https://github.com/harumiWeb/tree-sitter-vba/actions/workflows/ci.yml/badge.svg)](https://github.com/harumiWeb/tree-sitter-vba/actions/workflows/ci.yml)
+
+Tree-sitter grammars for Visual Basic for Applications (VBA) and Visual Basic 6
+(VB6), targeting exported Excel/VBA source files such as `.bas`, `.cls`, and
+`.frm`, and VB6 project sources including `.frm` forms and `.ctl` user
+controls. Two parsers, `vba` and `vb6`, are generated from one shared grammar
+core.
+
+[Try tree-sitter-vba online](https://harumiweb.github.io/tree-sitter-vba/)
+
+The playground runs the `vba` parser through WebAssembly entirely in your
+browser. Use it to test parser compatibility, inspect synchronized concrete
+syntax trees, and see the official `highlights.scm` query in action; source
+code is not uploaded or sent to a parsing service.
+
+This grammar is designed as a parsing foundation for editor and tooling use
+cases, including syntax highlighting, folding, tags, outline extraction, symbol
+inspection, linting, formatting, and future LSP integrations.
+
+## Why another VBA grammar?
+
+Several tree-sitter grammars exist for Visual Basic-family languages, but VBA has its own syntax and practical edge cases, especially in code exported from the VBE.
+
+This project focuses on practical Excel/VBA source compatibility:
+
+- `.bas`, `.cls`, and `.frm` files exported from the VBE
+- real-world VBA fixtures
+- UserForm metadata
+- conditional compilation
+- error handling
+- Excel-style member access and procedure calls
+- editor queries for highlights, folds, and tags
+- permissive MIT licensing
+
+VB6 shares its statement grammar with VBA, so the `vb6` parser reuses all of
+the above and adds what is VB6's own: the `.frm`/`.ctl` form header and `.cls`
+class header as the VB6 IDE writes them, and the constructs that predate VBA.
+
+## Status
+
+This is a `v0.x` public release.
+
+The grammar is already usable for syntax-aware tooling such as highlighting,
+folding, tags, outline extraction, and initial symbol analysis. The current test
+suite covers 261 focused VBA corpus cases, 45 VB6 corpus cases, generated
+`Select Case` stress coverage through 500 clauses, and 481 checked-in VBA
+example files without `ERROR` or `MISSING` recovery nodes. The `vb6` parser is
+additionally measured against 1,436 real VB6 source files from 71 public
+projects; see [Testing](#testing) for the numbers.
+
+It is not yet a complete VBA grammar. Node names and tree shapes may still change before `v1.0.0`.
+
+## Two parsers from one grammar
+
+The statement grammar is written once, in `common/define-grammar.js`, as a
+function of the dialect. Each dialect directory holds a one-line entry point,
+its generated parser, and its corpus tests:
+
+```text
+common/define-grammar.js   the whole grammar: defineGrammar("vba" | "vb6")
+vba/grammar.js             module.exports = require("../common/define-grammar")("vba")
+vba/src/                   generated vba parser; node-types.json is tracked, parser.c is generated
+vba/test/corpus/           261 VBA corpus cases
+vb6/grammar.js             module.exports = require("../common/define-grammar")("vb6")
+vb6/src/scanner.c          a four-token external scanner (see Design principles)
+vb6/src/                   generated vb6 parser
+vb6/test/corpus/           45 VB6 corpus cases
+corpus/                    provenance of the VB6 acceptance corpus; the source files are fetched, not vendored
+examples/                  481 VBA example files
+```
+
+Everything dialect-specific is an `isVB6` / `isVBA` branch inside the core;
+`grep -n "isVB6\|isVBA" common/define-grammar.js` lists the complete delta.
+This is the layout `tree-sitter-typescript` uses for TypeScript and TSX, and
+the reason both grammars live in subdirectories: with more than one grammar in
+`tree-sitter.json`, the CLI picks the grammar from the working directory.
+
+Generated artifacts follow [ADR 0002](docs/adr/0002-generated-parser-artifacts.md):
+`src/parser.c` and `src/grammar.json` are not tracked in either dialect
+directory, `src/node-types.json` is, and `bindings/go/parser.c` is a tracked copy
+of the `vba` parser.
+
+## Installation
+
+```bash
+npm install tree-sitter tree-sitter-vba
+```
+
+The npm package's native addon is the `vba` parser. The `vb6` parser ships as
+generated C source (`vb6/src/parser.c` and `vb6/src/scanner.c`) for the
+tree-sitter CLI and for direct embedding; Node.js and Go bindings for it are
+not provided yet.
+
+## Usage
+
+```js
+const Parser = require("tree-sitter");
+const VBA = require("tree-sitter-vba");
+
+const parser = new Parser();
+parser.setLanguage(VBA);
+
+const tree = parser.parse(`
+Sub Hello()
+    Debug.Print "Hello"
+End Sub
+`);
+
+console.log(tree.rootNode.toString());
+```
+
+To parse VB6 from the command line, run the CLI inside the `vb6` directory,
+which is how it selects the grammar:
+
+```bash
+cd vb6 && tree-sitter parse /path/to/Form1.frm
+```
+
+`scripts/run-tree-sitter.mjs --cwd vb6 parse <file>` does the same from the
+repository root.
+
+## Go Usage
+
+The Go binding is self-contained when installed through Go modules; its package
+directory includes the generated C parser needed by cgo. It exposes the `vba`
+parser.
+
+```go
+package main
+
+import (
+	"fmt"
+
+	tree_sitter "github.com/tree-sitter/go-tree-sitter"
+	tree_sitter_vba "github.com/harumiWeb/tree-sitter-vba/bindings/go"
+)
+
+func main() {
+	parser := tree_sitter.NewParser()
+	defer parser.Close()
+
+	parser.SetLanguage(tree_sitter.NewLanguage(tree_sitter_vba.Language()))
+
+	tree := parser.Parse([]byte("Sub Hello()\nEnd Sub\n"), nil)
+	defer tree.Close()
+
+	fmt.Println(tree.RootNode().ToSexp())
+}
+```
+
+Example output:
+
+```text
+(source_file
+  (sub_declaration
+    name: (identifier)
+    (parameter_list)
+    body: (block
+      (call_statement
+        callee: (qualified_member_expression
+          receiver: (identifier)
+          operator: "."
+          member: (identifier))
+        arguments: (unparenthesized_argument_list
+          (string_literal))))))
+```
+
+## Node.js native build requirements
+
+The npm package currently builds its native addon from source during
+installation. Prebuilt native binaries may be added in a future release.
+
+A supported Python installation, a C/C++ toolchain, and the platform
+requirements documented by `node-gyp` are required.
+
+On Windows, Visual Studio 2022 Build Tools with the **Desktop development with
+C++** workload is recommended.
+
+At minimum, Windows users typically need:
+
+- Python 3.11 or later
+- Visual Studio 2022 Build Tools
+- MSVC C++ x64/x86 build tools
+- Windows 10 SDK or Windows 11 SDK
+
+If multiple Python installations are present, configure npm to use the intended
+Python executable:
+
+```powershell
+npm config set python "C:\Users\<you>\AppData\Local\Programs\Python\Python312\python.exe"
+```
+
+For reliable native builds on Windows, run installation from:
+
+```text
+x64 Native Tools Command Prompt for VS 2022
+```
+
+## Supported syntax
+
+The grammar currently supports:
+
+- apostrophe comments and `Rem` comments
+- string, integer, floating-point, boolean, date, `Nothing`, `Null`, and `Empty`
+  literals
+- decimal and hexadecimal literals with common VBA type characters, including
+  Currency (`@`) and LongLong (`^`), exponent notation such as `1E-3`, and
+  abbreviated decimal forms such as `.5` and `1.`
+- identifiers with common VBA type-declaration characters, including `@` and
+  `^`; in declarations the Single character `!` (`Dim X!`, `Const Y! = 1`) is
+  a `bang_identifier` because it shares its token with the bang member operator
+- identifiers, simple type clauses, dotted type names, and array type suffixes
+- `Attribute` statements
+- `Option Explicit`, `Option Private Module`, `Option Compare`, and `Option Base`
+- `Implements` statements
+- `Sub`, `Function`, and `Property Get/Let/Set` procedures
+- `Event` declarations
+- `RaiseEvent` statements
+- `Dim`, `Static`, `WithEvents`, visibility-based variable declarations,
+  arrays, `ReDim`, `Erase`, and `Const`
+- default type declaration statements such as `DefInt` and `DefStr`
+- `Type` and `Enum` declarations
+- external `Declare Function` and `Declare Sub` declarations, including
+  `PtrSafe`, `Lib`, and `Alias`
+- simple assignments and `Set` assignments
+- `Name oldPath As newPath` file rename statements
+- calls, named arguments, omitted arguments anywhere in an unparenthesized
+  list, call-site `ByVal`, member access, bang member access, and leading-dot
+  member access
+- `Debug.Print`, unparenthesized `Print` methods, and `? expr` Debug.Print
+  shorthand statements, including comma- and semicolon-separated output lists
+  with trailing output-position controls
+- `New` expressions and `As New` declarations
+- fixed-length string declarations
+- `AddressOf` expressions
+- common VBA operator precedence for arithmetic, concatenation, comparison, and
+  logical operators; `=`, `<>`, `<`, `<=`, `>`, `>=`, `Is`, and `Like`
+  comparisons are represented as `comparison_expression`
+- comparison expressions in `Case` clauses, including object identity forms
+  such as `Case obj Is Nothing` and condition lists such as `Case x = 1, y = 2`
+- block `If`, including colon-separated multi-statement single-line branches;
+  `Select Case`; and nested single-line `For`, `For Each`, `Do`, `While/Wend`,
+  and `With`
+- `Next` counter lists such as `Next i` and `Next j, i`
+- `On Error`, computed `On ... GoTo`/`GoSub`, `Resume`, `GoTo`, labels,
+  standalone `End`, and `Exit` statements
+- common file I/O statements: `Open`, `Input #`, `Line Input #`, `Print #`,
+  `Write #`, `Close`, `Get #`, `Put #`, `Lock`, `Unlock`, `Seek`, and `Reset`,
+  including Print/Write output lists with trailing `;` or `,` and common
+  `Access ... Shared` locking clauses
+- simple runtime statements: `Stop`, `Beep`, `Load`, and `Unload`
+- Access report `Line` drawing calls that use coordinate ranges such as
+  `Me.Line (x, y)-(x2, y2)`
+- `TypeOf ... Is ...` checks, including dotted type names such as `Access.Line`
+- numeric line labels, numbered statements, and numbered control-flow delimiters
+- conditional compilation with `#Const`, `#If`, `#ElseIf`, `#Else`, and
+  `#End If`, including statement branches inside procedures, members inside
+  `Type` and `Enum` declarations, and alternative procedure headers
+- line continuations, including trailing whitespace, and colon-separated
+  statements and `Enum` members
+- minimal `.frm` and `.cls` export metadata, including `VERSION`,
+  `Begin ... End`, `BeginProperty ... EndProperty`, GUID form blocks, and
+  `.frx` blob references
+- initial `highlights.scm`, `folds.scm`, and `tags.scm` queries
+
+### VB6 additions
+
+The `vb6` parser accepts everything above except the VBA7-only constructs
+(`PtrSafe`, `LongPtr`, `LongLong`, `DefLngPtr`, `DefLngLng`: `PtrSafe` on a
+`Declare` is an error, `LongPtr` and `LongLong` are ordinary user type names),
+and adds:
+
+- the `.frm`/`.ctl` header as the VB6 IDE writes it: `Object = "{GUID}#2.0#0"; "X.OCX"`
+  component references, nested `Begin Lib.Class Name ... End` blocks to any
+  depth, `BeginProperty` blocks with the OCX GUID and indexed names
+  (`BeginProperty ColumnHeader(1) {...}`), property values as strings, numbers,
+  negative numbers, hex literals, annotated booleans (`-1  'True`), `.frx`/`.ctx`
+  resource references with hexadecimal offsets and the `$` string prefix, menu
+  shortcuts (`^O`, `{F5}`, `+{DEL}`), indexed property-bag names
+  (`Tab(0).Control(1) = "txt(1)"`), and Single values written with a decimal
+  comma by IDEs on European locales (`FontSize = 8,25`); VB3-era
+  `Begin Form Form1` headers parse too
+- the `.cls` header, `VERSION 1.0 CLASS` / `BEGIN ... END`, and two-valued
+  `Attribute VB_Ext_KEY = "A" ,"B"` lines written by Class Builder
+- form and PictureBox graphics statements with coordinate pairs, bare or
+  `obj.`-qualified: `Line (x1, y1)-(x2, y2), color, BF`, `PSet (x, y), c`,
+  `Circle (x, y), r, , , , aspect`, and `Scale (0, 0)-(w, h)`, with any
+  trailing argument omitted
+- a comment whose line ends in ` _` continuing onto the next line, which
+  compiled VB6 projects rely on
+- multi-statement single-line loops, `For i = 1 To n: a = 1: b = 2: Next`
+- calls whose first argument is an implicit member, `Foo .Bar, x`, read as a
+  call with two arguments rather than a chain `Foo.Bar` with an omitted
+  argument, through the external scanner described under Design principles
+- constructs VBA also has but the base VBA grammar never accepted; most remain
+  `vb6` only so that the `vba` parser's trees and parse table stay within
+  the measured size and CST contract (see
+  [ADR 0005](docs/adr/0005-vba-parser-stays-the-base-grammar.md)).
+  Since v0.14.5, VBA also accepts `Let`, `On Local Error`, and `Global`;
+  the remaining `vb6`-only forms include `LSet`/`RSet`, `GoSub`/`Return`,
+  `Dim WithEvents`, `ReDim x(n) As T`, octal `&O` literals,
+  `AddressOf Module.Procedure`, `Name.Member` receivers, a comparison as the
+  left operand of another (`a = b <> 0`, `x Is Nothing = False`), a comparison
+  as the `Select Case` selector, `#If` around whole `Case` clauses, an empty
+  `Else` in a single-line `If`, `::` between inline
+  statements, a colon before a `For` body's line break, a `Type` member named
+  `End`, and the Single `!` suffix in expression positions (`total! = 0#`,
+  `rec.Percent! = 0#`)
+- an indexed or property target on the left of `=` is an assignment; the base
+  lets `arr(i) = x` fall to a call of `arr` with the comparison `(i) = x` as
+  its argument when tree shape happens to prefer it
+
+The VB6-only rules are gated by `isVB6` in the core and add these node types:
+`frm_object_reference`, `frm_shortcut_value`, `frm_locale_number`,
+`frm_property_index`, `frm_property_name`, `frm_blob_offset`, `line_statement`,
+`pset_statement`, `circle_statement`, and `scale_statement`. The rest of the
+additions in the list above are shared by both dialects; the `vba` tree shapes
+pinned by the VBA corpus are unchanged.
+
+## Declaration node API
+
+This release line is still pre-`1.0.0`, and declaration node shapes may change
+when doing so makes syntactic metadata more directly available to downstream
+tools.
+
+Declaration consumers should prefer these structural nodes and fields over
+source-text scanning:
+
+- `property_get_declaration`, `property_let_declaration`, and
+  `property_set_declaration` distinguish property accessors directly.
+- `declare_sub_statement` and `declare_function_statement` distinguish external
+  declaration kind directly.
+- declaration headers expose stable fields such as `visibility`, `name`,
+  `parameters`, `type`, `library`, `alias`, `ptrsafe_modifier`, `body`, and
+  `end` where applicable. Procedure terminators are exposed through
+  `end_sub_statement`, `end_function_statement`, and `end_property_statement`
+  nodes.
+- procedure declarations expose additional procedure modifiers through a
+  `modifiers` field, separate from `visibility`.
+- variable, constant, type-member, and parameter declarations expose stable
+  `name`, `bounds`, `type`, `initializer`, `passing_mode`, `optional_modifier`,
+  `paramarray_modifier`, and `default_value` fields where syntactically valid.
+- `with_events_modifier`, `static_modifier`, `byval_modifier`, `byref_modifier`,
+  `optional_modifier`, `paramarray_modifier`, and `ptrsafe_modifier` are
+  explicit modifier nodes.
+- `implements_statement` exposes its target as `name`, and
+  `attribute_statement` exposes `name` and `value`.
+- a name written with the Single type character is a `bang_identifier` whose
+  children are the `identifier` and the `!`; the other type characters stay
+  inside the `identifier` token. In `vba` the node occurs where a declaration
+  names something (`Dim`, `Const`, `ReDim`, `Type` members, parameters,
+  `Function` names); `vb6` also produces it in expression positions.
+
+## Expression node API
+
+Member access and calls expose stable fields for analysis tools:
+
+- `qualified_member_expression` uses `receiver`, `operator`, and `member`.
+- `implicit_member_expression` uses `operator` and `member`; it has no
+  `receiver`, matching leading-dot and leading-bang VBA syntax.
+- `call_expression` uses `function` and `arguments`, where `arguments` is an
+  `argument_list`.
+- `call_statement` uses `callee` and, when arguments are present, `arguments`.
+  Parenthesized arguments use `argument_list`; statement-style arguments use
+  `unparenthesized_argument_list`. Unparenthesized `Print` methods are the
+  Print-family exception: their `arguments` field contains an `output_list`.
+- `output_list` is shared by `Debug.Print`, `?`, `Print #`, `Write #`, and
+  unparenthesized `Print` methods. Output expressions use the `value` field;
+  `;` and `,` output-position controls use the `position` field containing a
+  `char_position` node. A position may follow the final output expression.
+- in the `vb6` parser, `line_statement` and `scale_statement` use `start` and
+  `end` (each a `coordinate_pair` with `x` and `y`), `pset_statement` uses
+  `point`, `circle_statement` uses `center` and `radius`, the qualified forms
+  expose the receiver through `method`, and trailing graphics arguments use
+  `argument`.
+- in the `vb6` parser, form header nodes expose `frm_property_statement` as
+  `name` and `value`, `frm_begin_block` as `type` and `name`,
+  `frm_begin_property_block` as `name`, `index`, and `guid`,
+  `frm_blob_reference` as `offset`, and `frm_object_reference` as `class` and
+  `file`.
+
+## Control-flow node API
+
+- `for_statement` and `for_each_statement` expose optional terminator counters
+  through `next_variables: next_variable_list`; `next_variable` is no longer
+  emitted.
+- `next_variable_list` always contains the ordered identifiers from `Next`,
+  whether there is one counter or several.
+- A colon-separated single-line `If` branch with multiple statements is an
+  `inline_statement_sequence`; one-statement branches keep their direct child.
+- `shared_next_for_body` represents an enclosing loop whose nested loop is
+  closed by the same `Next ...` counter list.
+- In the `vb6` parser, `single_line_block` owns the colon after each of its
+  statements, so an inline loop may hold several statements before its `Next`,
+  `Loop`, or `End With`. The `vba` parser keeps its current shape.
+
+## Known limitations
+
+This grammar parses VBA and VB6 syntax only.
+
+It does not currently provide:
+
+- type checking
+- semantic analysis
+- Excel Object Model or COM reference knowledge
+- validation of identifier, member, type, procedure, workbook, or reference
+  existence
+- validation of context-sensitive statement placement
+- a formatter
+- an LSP server
+- complete coverage of every VBA expression edge case
+- semantic interpretation of `.frm` designer metadata
+
+For example, invalid placement of `Exit For`, unresolved procedure calls, invalid
+Excel object members, or missing workbook references are intentionally left to
+downstream tools.
+
+General expression-level `=` comparison is still context-limited to avoid
+ambiguity with assignment.
+
+For the `vb6` parser:
+
+- Input must be UTF-8 (or UTF-16 through the C API). VB6 source is stored in the
+  machine's code page, and bytes above 0x7F fed in raw never match the
+  identifier class, so identifiers such as `BtnAñadir` or a Big5 library name
+  in a `Begin` line are errors until the file is transcoded. Comments and
+  strings survive raw bytes; identifiers do not. Nine of the 1,436 corpus files
+  fail this way without transcoding.
+- `Debug.Assert (a >= b And c <= d) Or e = 0`, a call statement whose single
+  unparenthesized argument is a parenthesized condition followed by `Or` and a
+  comparison, parses as `Debug.Assert(...) Or e` with an orphaned `= 0` and two
+  `MISSING` nodes. The same shape with plain identifiers inside the parentheses
+  parses correctly; the comparisons inside the parentheses tip a GLR fork. One
+  corpus file, one line.
+- `.vbp` project files, `.dsr` designers, and `.pag` property pages are not
+  parsed; `.frx`/`.ctx` binaries are referenced, not read. VBA7 constructs are
+  rejected, and `#If VBA7` blocks are left to the conditional-compilation rules.
+- There are no Node.js or Go bindings for the `vb6` parser yet; the npm addon,
+  the Go module, and the Wasm artifact are the `vba` parser.
+
+## Queries
+
+This package includes initial Tree-sitter queries for:
+
+```text
+queries/highlights.scm
+queries/folds.scm
+queries/tags.scm
+```
+
+These queries are intended as a starting point for editor integrations and
+tooling. They may evolve as the grammar stabilizes. Both grammars in
+`tree-sitter.json` point at the same query files; the VB6-only node types are
+not yet highlighted.
+
+## Development
+
+Install dependencies:
+
+```bash
+pnpm install
+```
+
+Generate both parsers, or one of them:
+
+```bash
+pnpm generate
+pnpm generate:vba
+pnpm generate:vb6
+```
+
+After grammar changes, keep the Go module artifact in sync:
+
+```bash
+cp vba/src/parser.c bindings/go/parser.c
+pnpm check:go-parser
+```
+
+Run corpus tests:
+
+```bash
+pnpm test
+```
+
+`pnpm test:corpus` runs the VBA corpus under the `vba` parser, the VB6 corpus
+under the `vb6` parser, and then the VBA corpus under the `vb6` parser through
+`scripts/test-shared-corpus.mjs`, which skips the cases whose expectations differ
+between the dialects by design and prints the reason for each. `pnpm
+test:corpus:vba` and `pnpm test:corpus:vb6` run one side.
+
+The tree-sitter CLI selects the grammar from the working directory, so run it
+inside `vba/` or `vb6/`; `scripts/run-tree-sitter.mjs --cwd vb6 <args>` does
+that from the repository root.
+
+Parse example files:
+
+```bash
+pnpm parse:examples
+```
+
+This recursively parses the checked-in VBA examples with the `vba` parser and
+fails if any parse tree contains an `ERROR` or `MISSING` node.
+
+Parse the VB6 acceptance corpus:
+
+```bash
+node scripts/fetch-corpus.mjs
+pnpm parse:corpus -- --transcode
+```
+
+See [Testing](#testing) for what the corpus is and how it is measured.
+
+Run queries against the example files:
+
+```bash
+pnpm query:examples
+```
+
+This validates that `highlights.scm`, `folds.scm`, and `tags.scm` can run
+against the checked-in examples.
+
+Run the coarse parser benchmark:
+
+```bash
+pnpm bench
+```
+
+This reports file counts, total bytes, parse time, node counts, and
+`ERROR`/`MISSING` counts for the checked-in examples. It is intended to catch
+large regressions, not to provide a strict microbenchmark.
+
+Run the full local check:
+
+```bash
+pnpm check
+```
+
+Build and test the standalone browser parser (the `vba` parser):
+
+```bash
+pnpm build:wasm
+pnpm test:wasm
+```
+
+Compare the native and browser parser targets on representative corpus and
+exported-module fixtures:
+
+```bash
+pnpm test:parity
+```
+
+The parity suite checks root node types, `ERROR` / `MISSING` recovery, and
+normalized syntax-tree structure. It uses the standalone Wasm parser through
+`web-tree-sitter`; the browser consumer smoke test separately covers loading
+and parsing in Chromium.
+
+The complete browser artifact, release, versioning, and loading contract is
+documented in [docs/specs/wasm-artifact.md](docs/specs/wasm-artifact.md).
+In brief, versioned releases attach `tree-sitter-vba.wasm` and its SHA-256
+checksum to the matching GitHub Release. The artifact is not committed or
+included in the npm package. A browser consumer uses the matching
+`web-tree-sitter@0.26.9` runtime:
+
+```js
+import { Language, Parser } from "web-tree-sitter";
+
+await Parser.init({ locateFile: () => "/assets/web-tree-sitter.wasm" });
+const language = await Language.load("/assets/tree-sitter-vba.wasm");
+const parser = new Parser();
+parser.setLanguage(language);
+const tree = parser.parse("Sub Hello()\nEnd Sub\n");
+try {
+  if (!tree) throw new Error("Parser did not return a syntax tree");
+  console.log(tree.rootNode.toString());
+} finally {
+  tree?.delete();
+  parser.delete();
+}
+```
+
+For a complete minimal static example, including local WebAssembly runtime
+assets and `ERROR` / `MISSING` reporting, see
+[examples/browser-consumer](examples/browser-consumer/README.md).
+
+## Testing
+
+Tree-sitter grammar behavior is tested with corpus files under:
+
+```text
+vba/test/corpus/
+vb6/test/corpus/
+```
+
+Community-reported production syntax regressions are kept permanently under
+`vba/test/corpus/regressions/`; each fixture records whether it was already
+supported or required a grammar change.
+
+When changing `common/define-grammar.js`, always add or update focused corpus
+tests in the dialect the change affects. Do not weaken existing expectations
+just to make a grammar change pass. A shared change must leave the `vba` trees
+unchanged; `scripts/test-shared-corpus.mjs` then runs the VBA corpus under the
+`vb6` parser as well, so a `vb6` divergence shows up as a failure there and is
+either fixed or recorded in the script's skip list with its reason. The 23
+skips today are the VBA7 constructs, the `.frx` offset node
+(`frm_blob_offset` in `vb6`, `number_literal` in `vba`), the pinned one-statement
+inline-loop shape, four cases where the `vba` parser emits an opaque
+`_ambiguous_call_statement` token that `vb6` parses with structure, and five
+error-recovery shapes on deliberately invalid input.
+
+A clean parse is not a stable tree. A bare call whose callee has been split
+into its own `expression_statement` still has zero `ERROR` nodes, so neither
+`pnpm parse:examples` nor the corpus can see that class of regression unless a
+case pins the exact shape. `scripts/compare-cst.mjs` closes that gap for the
+`vba` parser: it parses all 481 example files with a reference checkout and with
+this checkout, strips positions, and lists every differing file by the node
+types each hunk adds and removes. Point `--base` at a `main` worktree and name
+the intended differences with `--allow`:
+
+```text
+git worktree add /tmp/tree-sitter-vba-main main
+node scripts/compare-cst.mjs --base /tmp/tree-sitter-vba-main --allow bang_identifier
+```
+
+Run it before proposing any change to `common/define-grammar.js` that touches
+calls, arguments, or precedence, and explain every remaining hunk in the pull
+request.
+
+The repository also includes real-world exported VBA examples. These examples
+are parsed in CI to catch regressions against practical Excel/VBA code.
+
+Broken or incomplete examples can be kept under:
+
+```text
+examples/broken/
+```
+
+These fixtures are intentionally excluded from `pnpm parse:examples`. Add
+focused recovery expectations under `vba/test/corpus/recovery.txt` when the
+surrounding tree shape should remain stable.
+
+### VB6 acceptance corpus
+
+The `vb6` parser is measured against 1,436 real VB6 source files (647 `.frm`,
+382 `.bas`, 283 `.cls`, 124 `.ctl`) from 71 public GitHub projects, including
+code from the late 1990s. The files are not vendored: `corpus/MANIFEST.md` pins
+every project to a full commit SHA with its licence and file counts,
+`corpus/FILES.tsv` lists every file, and `scripts/fetch-corpus.mjs` clones each
+project at its pinned SHA and copies exactly those paths, so the set is
+byte-identical to the one measured:
+
+```bash
+node scripts/fetch-corpus.mjs          # all 71 projects
+node scripts/fetch-corpus.mjs --only <owner__repo>
+pnpm parse:corpus -- --transcode
+```
+
+Files are copied as raw bytes; most are Windows-1252 with CRLF and must not be
+re-encoded, so the runner transcodes at read time per `corpus/ENCODINGS.json`
+(180 files via Windows-1252, 56 via GBK, 22 via Big5). Keeping the source out of
+the tree also means this repository redistributes nobody else's code; five of
+the 71 projects are copyleft and are marked as such in the manifest.
+
+Measured on 2026-09-25 with tree-sitter CLI 0.26.9:
+
+| Check | Result |
+| --- | --- |
+| VB6 corpus, transcoded | 1,432 of 1,436 files parse with zero `ERROR` and zero `MISSING`. Three files carry syntax errors VB6 itself rejects and are listed with the defect in `corpus/EXCLUDED.json`; the runner reports them separately. One file keeps two `MISSING` nodes on the `Debug.Assert` shape described under Known limitations |
+| VB6 corpus, raw bytes | nine further files fail on non-ASCII identifiers in legacy code pages |
+| VBA corpus under `vba` | 261 of 261 cases; 481 of 481 example files without `ERROR` or `MISSING` |
+| VB6 corpus cases under `vb6` | 45 of 45 |
+| VBA corpus under `vb6` | 238 of 238 cases run, 23 skipped by design as listed above |
+| Parse time | the largest corpus file, a 2.2 MB module, parses in about 410 ms (5.4 MB/s); a 500 KB `.frm` in 97 ms |
+
+## Design principles
+
+This repository parses VBA and VB6 syntax only.
+
+It does not validate whether identifiers, types, members, procedures, workbook
+objects, or references are semantically valid. Those concerns belong in
+downstream tools such as `xlflow`, `xlflow-lsp`, editor extensions, or other
+analysis tools.
+
+Node names should remain stable once introduced because downstream query files
+and integrations may depend on them. However, because this is a `v0.x` release,
+node names and tree shapes may still change before `v1.0.0`.
+
+Over-acceptance is preferred to ambiguity: where a choice exists between
+rejecting invalid code and keeping the tree unambiguous, the grammar keeps the
+tree and leaves validation to downstream tools.
+
+### Why the `vb6` parser has an external scanner
+
+VB6 reads whitespace before a member operator at statement level: `Foo .Bar, x`
+is a call to `Foo` with two arguments, `Foo.Bar , x` a call to `Foo.Bar` with an
+omitted first argument. A context-free grammar never sees the space. The `vba`
+parser works around one shape with whole-line regex tokens
+(`_ambiguous_call_statement`), which keep the tree free of errors but hide the
+statement's structure, and `token.immediate(".")` cannot share a lexer state
+with a regular `"."`. The scanner in [vb6/src/scanner.c](vb6/src/scanner.c)
+emits `_dot_immediate` / `_dot_spaced` / `_bang_immediate` / `_bang_spaced`:
+member chains use the immediate kind only, an implicit member may start with
+either, a dot after a line continuation counts as immediate (`x("y") _`
+followed by `.Add(...)` is a chain), and a dot followed by a digit is left to
+the internal lexer (`.5`). It carries no state. The `vba` parser has no scanner.
+
+## Versioning
+
+Recommended interpretation of the current release line:
+
+- `0.1.x`: parser coverage fixes, query fixes, and non-breaking improvements
+- `0.2.x`: expanded real-world VBA coverage; tree shapes may still evolve
+- `0.x.0`: notable grammar expansion or tree-shape changes
+- `1.0.0`: node names and tree shapes are considered stable for downstream use
+
+## Related projects
+
+This grammar is intended to support practical VBA tooling, including the
+[harumiWeb/xlflow](https://github.com/harumiWeb/xlflow) ecosystem for AI-assisted Excel/VBA development.
+
+## License
+
+This project is licensed under the MIT License. See the [LICENSE](LICENSE) file for details.
+
+### Third-party examples
+
+The `examples/third_party` directory may contain third-party fixture files under
+their original licenses. These files are provided for parser coverage only and
+are not licensed under this repository's MIT license unless explicitly stated.
+See [THIRD_PARTY_LICENSE.md](THIRD_PARTY_LICENSE.md) for the project inventory,
+upstream sources, acknowledgements, and local license records.
+
+The VB6 acceptance corpus is not part of this repository; `corpus/MANIFEST.md`
+records the repository, commit, and licence of every project it is fetched
+from.