add-language-support

name	add-language-support
description	Implements mutation testing support for new programming languages using tree-sitter grammars and the LanguageEngine trait. Triggers on "add language support", "add [language] support", "implement [language]", "new language", or mentions of specific languages like Python, TypeScript, C++, Java.
license	Apache-2.0
metadata	{"author":"trailofbits","version":"1.1","mewt-version":">=2.0.0"}
allowed-tools	["Read","Write","Edit","Bash","Glob","AskUserQuestion"]

Adding Language Support to Mewt

Implements mutation testing support for new programming languages using tree-sitter grammars and the LanguageEngine trait.

When to Use

Use this skill when:

• Adding support for a new programming language (Python, JavaScript, Go, etc.)
• Implementing tree-sitter grammar integration
• User asks to "add language support" or "implement X language"
• Extending mewt's language capabilities

When NOT to Use

Don't use for using mewt: If the user wants to USE mewt for mutation testing (not add language support), use the mewt skill instead.

Don't use for grammar development: This assumes a tree-sitter grammar already exists. If no grammar exists, that's a separate specialized task beyond this skill's scope.

Essential Principles

Grammar First: Never attempt to write a tree-sitter grammar from scratch. Always find and verify an existing tree-sitter grammar repository before proceeding. Grammar development requires specialized expertise.

API Simplicity: The LanguageEngine trait has only 4 methods. Keep implementations minimal - no helper methods, just the trait interface and inline grammar loading.

Resolver First (Dialect-Aware Families): If a language family has multiple dialects sharing an extension, resolution must be owned by the registry/resolver layer (precedence, canonical labeling, deterministic ambiguity handling). Do not add command-specific branching for final language+dialect selection.

Canonical Labels Only: For dialect-aware families, use canonical selectors/labels (e.g., move, move/sui, move/iota) and avoid introducing legacy alias compatibility unless explicitly requested.

Common First: Start with COMMON_MUTATIONS only. Add language-specific mutations only for unique constructs not covered by common patterns. Most languages need zero custom mutations.

Per-Slug Mutation Tests: Every mutation slug exposed by the engine must have a dedicated test module under tests/<language>/mutations/<SLUG>.rs. The guard test in tests/languages.rs enforces this convention. New languages must be wired into the guard before landing.

Integration Conformance First: Every language tests/<language>/integration_tests.rs must run the shared conformance harness from tests/conformance.rs, then add language-specific integration assertions as needed.

Example Fixture Policy: Keep canonical example fixtures at tests/<language>/example.<ext> (JavaScript may keep multiple canonical fixtures: example.js, example.ts, example.jsx, example.tsx). Integration tests should treat these as smoke checks (!mutants.is_empty()), while per-slug tests should use inline fixtures for precise assertions.

Verification Required: Each phase has explicit exit criteria. Do not proceed to the next phase until all validation passes.

When to Use

• User explicitly requests adding support for a specific programming language
• User mentions "add language support" or "new language"
• User provides a tree-sitter grammar repository URL
• User asks how to extend mewt with new languages

When NOT to Use

• Language already supported (check src/languages/) - use existing implementation
• User wants to modify existing language support - use general code editing
• User wants to add mutations to existing language - edit src/languages/<lang>/mutations.rs
• No tree-sitter grammar exists - inform user and halt (cannot create grammars)

Linear Progression

Phase 1: Grammar Acquisition

Entry Criteria: User has requested language support

Actions:

1. Check if tree-sitter grammar repository URL is provided

   • If not, ask user with common patterns:
     • https://github.com/tree-sitter/tree-sitter-<language>
     • https://github.com/<maintainer>/tree-sitter-<language>
   • Offer to search the web if needed

2. Once URL confirmed, use Edit to add to grammars/update.sh (lines 15-23):

   declare -A REPO_URLS=(
     ["<language>"]="<tree-sitter-repo-url>"
   )
   
   declare -A GRAMMAR_PATHS=(
     ["<language>"]=""  # or subdirectory if nested
   )
   

3. Run grammar extraction:

   cd grammars && bash update.sh <language> false
   

Exit Criteria:

• Grammar files exist in grammars/<language>/src/
• Must have: parser.c, tree_sitter/ directory
• May have: scanner.c, node-types.json

Phase 2: Build System Integration

Entry Criteria: Grammar files extracted successfully

Actions:

1. Use Edit to add build configuration to build.rs after line 57:

   // Build <Language> grammar
   let <language>_dir: PathBuf = ["grammars", "<language>", "src"].iter().collect();
   build_grammar(&<language>_dir, "tree-sitter-<language>");
   

2. Verify compilation:

   cargo check
   

Exit Criteria:

• cargo check completes without errors
• Grammar library compiles successfully

Phase 3: Language Engine Implementation

Entry Criteria: Grammar builds successfully

Actions:

1. Create language module structure:

   mkdir -p src/languages/<language>
   

2. Write module declaration (src/languages/<language>/mod.rs):

   pub mod engine;
   pub mod mutations;
   pub mod syntax;
   

3. Create syntax mappings (src/languages/<language>/syntax.rs):

   • Use Read to examine grammars/<language>/src/node-types.json
   • Map node and field names from the grammar

   pub mod nodes {
       pub const IF_STATEMENT: &str = "if_statement";
       pub const RETURN_STATEMENT: &str = "return_statement";
       // Add more from node-types.json
   }
   
   pub mod fields {
       pub const CONDITION: &str = "condition";
       pub const ARGUMENTS: &str = "arguments";
   }
   

   • For every operator-focused slug (AOS, AAOS, BOS, BAOS, LOS, SAOS, COS), confirm which node kinds the grammar uses for those operators (e.g., augmented_assignment_expression versus binary_expression). Only plug a node kind into patterns::shuffle_* after verifying it in node-types.json, and include any language-specific operator tokens (such as Go's &^= or JavaScript's **=/>>>=).

4. Create mutations file (src/languages/<language>/mutations.rs):

   use crate::types::Mutation;
   pub const <LANGUAGE>_MUTATIONS: &[Mutation] = &[];
   

5. Implement engine (src/languages/<language>/engine.rs):

   • Use Read on src/languages/rust/engine.rs as reference
   • Follow the 4-method trait pattern:

   use std::sync::OnceLock;
   use tree_sitter::Language as TsLanguage;
   
   use crate::LanguageEngine;
   use crate::mutations::COMMON_MUTATIONS;
   use crate::patterns;
   use crate::types::{Mutant, Mutation, Target};
   use crate::utils::{node_text, parse_source};
   
   use super::mutations::<LANGUAGE>_MUTATIONS;
   use super::syntax::{fields, nodes};
   
   static <LANGUAGE>_LANGUAGE: OnceLock<TsLanguage> = OnceLock::new();
   
   unsafe extern "C" {
       fn tree_sitter_<language>() -> *const tree_sitter::ffi::TSLanguage;
   }
   
   pub struct <Language>LanguageEngine {
       mutations: Vec<Mutation>,
   }
   
   impl <Language>LanguageEngine {
       pub fn new() -> Self {
           let mut mutations: Vec<Mutation> = Vec::new();
           mutations.extend_from_slice(COMMON_MUTATIONS);
           mutations.extend_from_slice(<LANGUAGE>_MUTATIONS);
           Self { mutations }
       }
   }
   
   impl LanguageEngine for <Language>LanguageEngine {
       fn name(&self) -> &'static str {
           "<Language>"
       }
   
       fn extensions(&self) -> &[&'static str] {
           &["<ext>"]
       }
   
       fn get_mutations(&self) -> &[Mutation] {
           &self.mutations
       }
   
       fn mutate(&self, target: &Target) -> Vec<Mutant> {
           let source = &target.text;
           
           // Load grammar once and cache it
           let language = <LANGUAGE>_LANGUAGE
               .get_or_init(|| unsafe { TsLanguage::from_raw(tree_sitter_<language>()) });
           
           let tree = match parse_source(source, language) {
               Some(t) => t,
               None => return Vec::new(),
           };
           let root = tree.root_node();
   
           let mut all_mutants = Vec::new();
           for m in &self.mutations {
               match m.slug {
                   "ER" => {
                       all_mutants.extend(
                           patterns::replace(
                               root,
                               source,
                               &[nodes::EXPRESSION_STATEMENT, nodes::RETURN_STATEMENT],
                               "panic!(\"mewt\")",  // Use language-appropriate error
                               &|node, src| !node_text(node, src).contains("panic!"),
                           )
                           .into_iter()
                           .map(|p| Mutant::from_partial(p, target, "ER")),
                       )
                   }
                   "IF" => {
                       all_mutants.extend(
                           patterns::replace_condition(
                               root,
                               source,
                               nodes::IF_STATEMENT,
                               fields::CONDITION,
                               &["if"],
                               "false",
                           )
                           .into_iter()
                           .map(|p| Mutant::from_partial(p, target, "IF")),
                       )
                   }
                   // Add more mutation patterns - see src/languages/rust/engine.rs
                   _ => {}
               }
           }
           all_mutants
       }
   }
   
   impl Default for <Language>LanguageEngine {
       fn default() -> Self {
           Self::new()
       }
   }
   

Exit Criteria:

• All 4 trait methods implemented
• Grammar loaded inline in mutate() (no helper methods)
• At least ER and IF mutations implemented
• Code compiles with cargo check

Phase 4: Language Registration

Dialect-aware note: If the language can have multiple dialects under one extension, also wire/update registry resolver metadata and canonical labeling rules so command flows consume one resolved selection object.

Entry Criteria: Engine implementation compiles

Actions:

1. Use Edit to add module to src/languages/mod.rs:

   pub mod <language>;
   

2. Use Edit to register in src/main.rs (find the LanguageRegistry section):

   registry.register(mewt::languages::<language>::engine::<Language>LanguageEngine::new());
   

Exit Criteria:

• Module exported in src/languages/mod.rs
• Engine registered in src/main.rs
• cargo build --release succeeds

Phase 5: Tests and Examples

For dialect-aware families that share an extension, add:

• Resolver integration tests proving deterministic selection for ambiguous extension paths.
• Dialect-divergence tests with at least one construct accepted by one dialect and rejected (or treated differently) by another.
• Filter/query tests using canonical labels/selectors (no implicit legacy alias guarantees).

Entry Criteria: Language builds and registers

Actions:

1. Scaffold the test directories:

   mkdir -p tests/<language>/mutations
   

   The mutation folder must contain one Rust module per slug (for example tests/<language>/mutations/AOS.rs).

2. Write canonical example fixture(s):

   • Most languages: tests/<language>/example.<ext>.
   • JavaScript-family languages: one canonical fixture per supported extension (example.js, example.ts, example.jsx, example.tsx) as needed.
   • Keep examples syntactically valid and representative, but small enough to stay smoke-test friendly.

3. Wire the language into the integration test entrypoint (tests/languages.rs):

   mod <language>;
   

4. Create tests/<language>/mod.rs to expose both integration and per-slug suites:

   mod integration_tests;
   mod mutations;
   

5. Author tests/<language>/mutations/mod.rs that re-exports every slug module:

   #![allow(non_snake_case)]
   
   #[path = "AAOS.rs"]
   mod aaos;
   #[path = "AOS.rs"]
   mod aos;
   // Repeat for every slug returned by engine.get_mutations()
   

   • Use uppercase filenames that match the slug (<SLUG>.rs) and map to lowercase module names via #[path = ...] where needed.
   • Keep manual slug wiring in mutations/mod.rs in sync with files on disk.

6. For each slug surfaced by engine.get_mutations(), create tests/<language>/mutations/<SLUG>.rs:

   • Prefer inline source fixtures for precise, slug-focused assertions.
   • Use integration helpers (create_test_target, shared slug assertion helpers) rather than duplicating setup code.
   • For operator families (AAOS/BAOS/SAOS/AOS/BOS/COS/LOS), cover all supported operators and include negative cases where useful.

7. Create tests/<language>/integration_tests.rs with this pattern:

   • Define a thin create_test_target(...) wrapper that delegates to tests/utils.rs.
   • Run conformance::run_common_language_checks(...) for baseline behavior.
   • Add canonical example-file smoke tests (!mutants.is_empty()).
   • Add only language-specific integration assertions beyond conformance (for example parser edge cases).

8. Update the guard in tests/languages.rs so the new language participates in per-slug coverage checks (import the engine and add a check_language(...) call).

Exit Criteria:

• Canonical example fixture(s) created at tests/<language>/example.<ext> (or JavaScript example.* set)
• Integration and per-slug test modules compiled into tests/languages.rs
• (Dialect-aware only) Resolver tests cover deterministic extension resolution and canonical dialect labeling
• (Dialect-aware only) Divergence tests assert meaningful behavior differences between dialects
• tests/<language>/integration_tests.rs runs run_common_language_checks(...)
• Every slug exposed by the engine has a dedicated test module under tests/<language>/mutations
• Guard test passes without missing or unexpected modules
• cargo test (or just test) passes the full suite

Phase 6: Validation

Entry Criteria: All tests pass

Actions:

1. Build release binary:

   cargo build --release
   

2. Verify mutations are registered:

   ./target/release/mewt print mutations --language <language>
   

   • Should list COMMON_MUTATIONS (ER, CR, IF, etc.)

3. Generate and verify mutants:

   ./target/release/mewt print mutants --target tests/<language>/example.<ext>
   

   • Verify reasonable number of mutants
   • Check mutations are diverse (ER, IF, CR, etc.)
   • Verify line numbers are accurate

4. Run full test suite:

   just test
   

Exit Criteria:

• cargo build --release succeeds
• mewt print mutations --language <language> lists mutations
• mewt print mutants generates mutants from example file
• All tests pass with just test
• No warnings with cargo check

API Quick Reference

The 4-Method Trait

pub trait LanguageEngine: Send + Sync {
    fn name(&self) -> &'static str;              // Language name
    fn extensions(&self) -> &[&'static str];     // File extensions
    fn get_mutations(&self) -> &[Mutation];      // Available mutations
    fn mutate(&self, target: &Target) -> Vec<Mutant>;  // Generate mutants
}

Key changes from older versions:

• ✅ mutate() replaces apply_all_mutations()
• ✅ No tree_sitter_language() method - load grammar inline
• ✅ No helper methods - keep implementations minimal

Common Mutation Patterns

Use Read on src/languages/rust/engine.rs for complete examples of:

Slug	Pattern	Purpose
ER	patterns::replace()	Replace statements with errors
CR	patterns::replace()	Replace statements with comments
IF	patterns::replace_condition()	Replace if conditions with false
IT	patterns::replace_condition()	Replace if conditions with true
WF	patterns::replace_condition()	Replace while conditions with false
LC	patterns::swap_branches()	Swap true/false branches
BL	patterns::replace()	Replace boolean literals
AOS	patterns::replace_operator()	Replace arithmetic operators
BOS	patterns::replace_operator()	Replace bitwise operators
LOS	patterns::replace_operator()	Replace logical operators
COS	patterns::replace_operator()	Replace comparison operators

Common Pitfalls

Node Type Mismatches

Problem: Using documentation names instead of actual grammar names
Solution: Always verify in grammars/<language>/src/node-types.json

FFI Function Naming

Problem: Incorrect external function name
Solution: Must be exactly tree_sitter_<language> (all lowercase, underscores for hyphens)

Missing Scanner

Problem: Build fails during C compilation
Solution: Some grammars need scanner.c - the update script handles this automatically

Over-Engineering

Problem: Adding helper methods, custom parsing logic
Solution: Keep it minimal - just implement the 4 trait methods

Too Many Custom Mutations

Problem: Adding language-specific mutations before verifying common ones work
Solution: Start with COMMON_MUTATIONS only. Most languages need zero custom mutations.

Compound Assignment Node Kinds

Problem: Assuming compound assignments share the binary_expression node kind
Solution: Check node-types.json for the exact node kind (e.g., augmented_assignment_expression, compound_assignment_expr) and wire AAOS/BAOS/SAOS to that. Include all language-specific operator tokens when configuring patterns::shuffle_operators.

Success Checklist

Dialect-Aware Extension Checklist (apply when relevant)

• Resolver owns final language+dialect selection (no command-level branching)
• Canonical dialect labels/selectors are documented and used consistently
• Ambiguous extension handling is deterministic and test-covered
• Dialect-divergence behavior is explicit and test-enforced
• Store/filter behavior uses canonical labels without assuming legacy alias support

Success Checklist

• Grammar files extracted to grammars/<language>/src/
• Build system updated in build.rs
• Module structure created (mod.rs, syntax.rs, mutations.rs, engine.rs)
• Syntax mappings verified from node-types.json
• Engine implements exactly 4 trait methods (no extras)
• Grammar loaded inline in mutate() using OnceLock
• Language registered in src/languages/mod.rs and src/main.rs
• Example files created with diverse syntax
• Integration tests written and passing
• cargo build --release succeeds without warnings
• mewt print mutations --language <language> works
• mewt print mutants generates mutants from examples
• All tests pass with just test

Example: Adding Go Support

Condensed walkthrough assuming tree-sitter-go URL is known:

# Phase 1: Edit grammars/update.sh to add go configuration, then:
cd grammars && bash update.sh go false

# Phase 2: Edit build.rs to add build_grammar call for go

# Phase 3: Create language implementation
mkdir -p src/languages/go
# Create mod.rs, syntax.rs, mutations.rs, engine.rs

# Phase 4: Edit src/languages/mod.rs to export go module
# Edit src/main.rs to register GoLanguageEngine

# Phase 5: Create tests
mkdir -p tests/go/examples
# Create example.go and integration_tests.rs

# Phase 6: Validate
cargo build --release
./target/release/mewt print mutations --language go
./target/release/mewt print mutants --target tests/go/example.go
just test