Changelog

All notable changes to this project are documented here. Format based on Keep a Changelog, adhering to Semantic Versioning.

[0.7.0] - 2026-03-28

calculate_xray_properties now returns Vector{XRayResult} instead of Dict{String, XRayResult}. Use index-based access (results[1]) instead of key-based (results["SiO2"]). This preserves duplicate formulas with different densities.
Batch validation is strict: all formulas are checked upfront and invalid ones are listed in a single ArgumentError before any computation begins.
calculate_single_material_properties now sorts energies ascending, matching batch API behavior.

Recursive descent formula parser supporting parenthesized groups: Ca(OH)2, Ca3(PO4)2, Fe((OH)2)3
Comprehensive debug logging: 27 @debug log points across 6 categories (:parser, :cache, :io, :validation, :computation, :batch). Enable with ENV["JULIA_DEBUG"] = "XRayLabTool".
ASCII validation in formula parser and element interpolator loader
Thread-safety test for concurrent cache access
Negative delta guard: critical_angle returns 0.0 instead of throwing DomainError when delta is negative near absorption edges

Thread-safe cache reads: replaced lock-free Dict.get() with locked reads (Julia's Dict is not safe for concurrent read + write)
Physical constants updated to CODATA 2018 / 2019 SI exact values (Planck, elementary charge, Avogadro — ~2e-7 relative change)
Formula parser no longer silently produces wrong results for parenthesized formulas

408 tests (up from 357): formula parsing, batch validation, thread safety, negative delta, logging, performance
Logging stdlib added to dependencies

Removed CSV.jl and DataFrames.jl dependencies (replaced with zero-dependency file parsing)
Internal F1F2_TABLE_CACHE replaced with INTERPOLATOR_CACHE (caches interpolator pairs directly)
Removed internal functions: load_scattering_factor_table, pchip_interpolators, create_interpolators

4.5x faster precompilation (62 to 24 transitive packages)
3.6x faster single-material calculations (cached interpolators, fused computation loops)
4x fewer allocations per call (19 KB vs 80 KB for 191 energy points)
Lock-free cache reads on warm-cache path (zero allocation on cache hits)
@threads :dynamic scheduling for better load balancing in batch mode
Non-allocating energy range validation
Type-stable hot loop via concrete ElementInterpolatorData type alias

Data race: element caches now pre-populated before @threads loop entry
Fluorine data file (f.nff) typo: first energy was 129.3 instead of 29.3
Energy range validation added to single-material path (previously only batch path validated)
Circular deprecation architecture: @deprecate macros replaced with explicit depwarn wrappers

Base.propertynames override for tab-completion on XRayResult
Documenter.jl documentation site with physics background, API reference, and migration guide
Consolidated CI pipeline with multi-threaded testing, formatting checks, and documentation deployment
357 tests (up from ~80): physical constants, numerical sanity, scaling properties, physics cross-validation, property-based formula parsing, performance regression guards

Dict-based getproperty with O(1) lookup replaces O(n) if-elseif chain
Narrowed imports: using Unitful: ustrip, using Base.Threads: @threads
SemVer-compatible compat bounds in Project.toml
Batch results collected via lock-free Vector indexed writes instead of Dict + ReentrantLock

New descriptive API:
- calculate_xray_properties replaces Refrac (multiple materials)
- calculate_single_material_properties replaces SubRefrac (single material)
New snake_case field names on XRayResult (see table below)
Backward-compatible getproperty override: old field names still work
Multi-threaded batch processing via @threads
Comprehensive docstrings on all public and internal functions

Note: f1 and f2 are unchanged.