feat(extract): MQL5 (.mq5/.mqh) extractor

[Dyrokk]

What

Adds MQL5 (.mq5/.mqh) extraction — MetaTrader 5 expert advisors, indicators and include headers.

The interesting part is what this doesn't do: MQL5 is C++ plus a handful of trading-specific constructs, so rather than a bespoke extractor it masks the four MQL5-only forms tree-sitter-cpp cannot parse and runs the existing _CPP_CONFIG pipeline over the result, using _extract_generic's source_override. That gets real classes, structs, functions, calls and includes for ~200 lines of new code instead of a second grammar.

The masked constructs:

Construct	Example
input / sinput / extern storage classes	input double Lots = 0.1;
input group "..." inspector heading	input group "=== GENERAL ==="
color literals	C'255,128,0'
datetime literals	D'2024.01.31 22:00'

Masking blanks bytes in place rather than deleting them, so every byte offset — and therefore every source_location — still matches the original file. Shifting them would silently point every node at the wrong line, which is the kind of bug that is invisible until someone clicks a link.

Inputs

Masking deliberately reduces input double Lots = 0.1; to an ordinary global, so input/extern declarations are recovered separately from the original source and re-attached as nodes, with references edges from the functions that read them.

This is the same call made for Pine in #2480, for the same reason: in an expert advisor the inputs are the strategy's tunable surface, and "which function reads magic_number" is the question people actually ask of one. The declared type is kept as mql5_input_type.

Attribution walks brace depth rather than parsing a second tree — exact here because MQL5 functions are file scope and never nested. String bodies are blanked before counting so a brace inside a literal cannot desynchronise it, and MetaEditor's house style (opening brace on the line after the signature) is handled explicitly, as is a single-line body.

Also: #property version/copyright/link land as metadata on the file node, and .mq5/.mqh join the C++ member-call resolver so CTrade-style object calls resolve.

Validation

Measured on a private corpus of live expert advisors, ~9,000 lines across 6 .mq5 files and 3 .mqh headers:

Parse errors: 196 → 0. Per file, before → after masking: 74→0, 61→0, 53→0, 4→0, 4→0, 0→0. The .mqh headers already parsed clean.

Extraction on the largest EA (3,051 lines): 133 nodes, 267 edges, 64 inputs recovered. Input reads were verified by hand against the source — every references edge for magic_number lands on a function that genuinely reads it (lines 261, 346, 621, 736, 783, 822, 869, 889, 1649), and no intra-file edge dangles.

Tests

tests/fixtures/sample.mq5 + tests/test_mql5.py, 13 cases covering masking (byte-offset preservation, each masked construct), extraction (functions, structs, inputs with their declared types, includes), attribution (enclosing function, single-line body, string-brace desync), the #property header, and a no-dangling-edges guard. One test asserts the fixture parses with zero tree-sitter errors after masking, so a future grammar bump that breaks this shows up immediately.

Full suite on this branch: 3904 passed. The 56 failures on my machine (test_skillgen, test_watch, test_uninstall_scope) are pre-existing on Windows — suite run with -p no:randomly before and after, failure set byte-identical.

Notes

• No new dependency: tree-sitter-cpp is already a core dep.
• Independent of feat(extract): Pine Script (.pine) extractor #2480 (Pine); this branch is cut from v8 and the two touch different lines. Either can merge first.

[graphify-labs]

Looks safe to merge — no coupling regressions and no blocking issues, checked against the code graph (not a self-assessment).

Graphify reviewed this change.

Looks safe to merge — no coupling regressions and no blocking issues, checked against the code graph (not a self-assessment).

---

Graphify review — findings

This PR adds MQL5 (.mq5/.mqh) file support to the extraction pipeline. It introduces a new graphify/extractors/mql5.py module and an extract_mql5 function that masks MQL5-only syntax the C++ grammar can't parse, then runs the existing C++ extraction pipeline over the result and separately recovers input/extern declarations plus #property metadata. Supporting changes register the new extensions across detection, analysis, build edge-family maps, the C++ member-call resolver, the dispatch table, README, and CHANGELOG, and a new test file (tests_test_mql5*) is included. The bulk of the remaining changed symbols appear to be renumbered rationale/internal helper identifiers, so the reviewer may want to confirm whether those are substantive changes or incidental renumbering.

Worth a look

• decode uses errors='replace' for input fact scanning while masking uses raw bytes, causing offset drift — graphify/extract.py:1719 · Escalate · medium
  • agreed by 2 of 2 members but NOT verified (no proof, no reproducing execution) — consensus is not a verdict; needs human review

Analysis details — impact, health, verification

Impact & health

Graphify review

Impact — 2462 functions depend on the 624 functions this change touches.

Health — this change adds coupling hotspots:

• new: extract_mql5() — 4 callers, 3 callees

Verification — 2462 functions in the blast radius were not formally verified this run (proofs are advisory here).

Gate & verification

graphify gate

PASS — objectively clean (no health regressions, tests not run — proofs not run this pass (advisory)). Grounded, not self-assessed.

Advisory (not blocking):

• verification_scope: 2204 function(s) in the blast radius were not formally verified this run

· 1 grounded finding(s) anchored inline below.

[graphify-labs]

⚠️ Health regression — extract_mql5()

high coupling complexity (Ca·Ce = 12).

Grounded coupling-delta finding (deterministic), not an LLM guess.