0.8.0 (2026-03-30)

Breaking Changes

Codebase quality improvements:

• Fix cu_benchmarks test crash by checking for SBF binary before loading mollusk
• Mark typed_builder re-export as #[doc(hidden)] non-stable API
• Add 11 tests for pina_cli error type Display impls
• Add cargo doc API docs check to verify:docs CI
• Rename loaders.rs → impls.rs for clarity
• Improve SAFETY documentation for all unsafe blocks in impls.rs

Features

Add multi-file module resolution to the IDL parser.

parse_program() now follows mod declarations from src/lib.rs to discover and parse additional source files. This enables IDL generation for programs that split code across multiple modules (e.g. src/state.rs, src/instructions/mod.rs).

New module: crates/pina_cli/src/parse/module_resolver.rs with 5 unit tests covering single-file crates, child modules, mod.rs style, missing modules, and inline modules.

The existing assemble_program_ir() function is preserved for backward compatibility and now delegates to the new assemble_program_ir_multi().

Implement opcode-aware CU cost model for the static profiler.

The profiler now decodes each 8-byte SBF instruction's opcode and assigns costs based on the instruction class:

• Regular instructions (ALU, memory, branch): 1 CU each
• Syscall instructions (call imm with src_reg=0): 100 CU each

Per-function profiles now include syscall_count and the text output shows a Syscall column. The JSON output includes total_syscalls and per-function syscall_count.

This replaces the previous flat 1-CU-per-instruction model which could underestimate programs with heavy syscall usage by 10-100x.

Fixes

Comprehensive documentation update across workspace.

New mdt providers (template.t.md):

• pinaCliCommands — CLI command reference table
• pinaIntrospectionDescription — introspection module overview
• pinaProfileDescription — static CU profiler overview

Updated documentation:

• docs/src/crates-and-features.md — added pina_profile, CLI commands table, multi-file parser note, pod arithmetic, codama renderer module structure
• docs/src/core-concepts.md — added Pod types table, arithmetic description, introspection section; fixed stale loaders.rs → impls.rs reference
• readme.md — added Pod arithmetic examples, static CU profiler section, replaced outdated 3-crate table with full workspace packages table
• crates/pina_cli/readme.md — added pina profile command, multi-file note
• Fixed missing CU_PER_INSTRUCTION import in profiler tests

mdt provider/consumer counts: 23/46 → 26/56.

Add 12 integration tests for pina::introspection module (previously 0% coverage).

Tests construct fake Instructions sysvar account data following the exact binary layout that pinocchio's Instructions parser expects, then exercise each introspection function end-to-end:

• get_instruction_count: single and multiple instructions
• get_current_instruction_index: correct index returned
• assert_no_cpi: passes for top-level, fails for CPI, checks correct index
• has_instruction_before: finds earlier programs, returns false when first
• has_instruction_after: finds later programs, returns false when last
• Instructions with account metas and data
• Wrong sysvar address rejection

Refactor pina_codama_renderer: split monolithic lib.rs into focused render modules.

• render/helpers.rs — string utilities, docs rendering, numeric casts
• render/discriminator.rs — discriminator type/value resolution
• render/types.rs — POD type rendering and defined-type pages
• render/accounts.rs — account struct, PDA helpers, accounts mod
• render/instructions.rs — instruction struct, account metas, data struct
• render/seeds.rs — variable and constant PDA seed expression rendering
• render/errors.rs — error enum pages and errors mod
• render/scaffold.rs — crate scaffold creation and file writing
• render/mods.rs — root mod and programs mod rendering

lib.rs retains only the public API (RenderConfig, read_root_node, render_idl_file, render_root_node, render_program) and the orchestrator render_program_to_files.

All 13 existing tests continue to pass.