Roadmap¶

VBAlidator was built in seven shippable PRs against the "Premium-Prechecker für KI-generierten VBA-Code" plan. This page tracks what's in, what's deferred, and what's next.

Done¶

PR	Phase	Highlights
#1	0 — Foundation	Single-line-If parser bug fix, lexer MISMATCH hardening, pytest migration, CI matrix (3 OS × 5 Py)
#2	1 — Control-flow	Real `For` / `Do` / `While` / `Select` body parsing, `ReDim` / `Erase` validation, lexer legacy-token support (`$`/`%`/`@` suffixes, `[bracket]` foreign names), 80-entry `std_model` backfill
#3	2.1 / 2.2 / 2.3	Label registry + jump validation, Set-vs-Let on bare-identifier LHS, Property Get/Let/Set arity & semantic checks
#4	2.4 / 2.5 / 2.7 / 2.8 / 2.9	Operator-type literal check, Const-expression validation, structural date-literal parser, fixed-length-string scope check, `DefInt`/`DefStr`/… implicit typing
#5	3.1 / 3.2 / 3.3 / 3.4 / 3.6	`Implements`-interface contract, `Event` / `RaiseEvent` / `WithEvents`, `PtrSafe` requirement, Enum-member uniqueness, `Option Explicit` warning
#6	4.1 / 4.2 / 4.3 / 4.4	Bundled host models (Excel/Word/Access/Outlook), 0–100 confidence score, JSON v2 schema, `from vbalidator import precheck` Python API
#7	4.5 / 4.6	Round-trip via Office COM, rule registry + auto-generated catalogue (`docs/rules/`)
#8	5	Maximum CI/CD: semantic-release → PyPI Trusted Publisher, multi-arch Docker → GHCR, security pipeline, MkDocs → GitHub Pages, repo hygiene

Deferred — high-value, queued for next iteration¶

P3.5 — Module-Level vs. Procedure-Level Statement Placement. Reject e.g. Type … End Type inside a Sub. Needs a parser-level refactor to track the active container.
P2.6 — UDT / Class member-chain depth. Today member chains validate the first hop fully and degrade to permissive Variant past the second .. Closing this requires a proper type system on the walker.
VBA350 — End placement in Function-returns. AI generators sometimes emit End Sub to close a Function. Easy rule, just needs a fixture.

Deferred — research / nice-to-have¶

Auto-fix engine. The catalogue's fix_hint field is the seed. Most useful for: missing Set, missing PtrSafe, wrong Property arity, Levenshtein-suggesting typo'd built-in names.
VS Code extension. LSP server wrapping precheck() for inline diagnostics on .bas / .cls / .frm files. Lives in a separate repo; the release workflow can dispatch a repository_dispatch event to trigger its build.
Web demo. Streamlit / Gradio app — paste code, get score.

Scoring philosophy¶

A score reaches 100 only when zero issues are present. Each severity subtracts a fixed weight from a starting 100:

score = max(0, 100 − Σ(weight × count))
weights: error=20  warning=3  info=1  compile_verified=30

The intent is: a single hard error always drops the score below the default 90 % CI gate. Style warnings (e.g. missing Option Explicit) nudge the score down by 3 each so a clean module landing at 100 genuinely means both compile-safe and style-clean.

Versioning¶

Conventional-Commit-driven semver via python-semantic-release:

Commit prefix	Version bump
`feat:`	minor
`fix:`, `perf:`, `refactor:`	patch
Body footer `BREAKING CHANGE:`	major
`docs:`, `test:`, `chore:`, `ci:`, `build:`, `style:`	none

Stable rule IDs are part of the public contract — they never change across releases. New rules get new IDs; deprecations are documented in the changelog before removal.