Release v7.1.8 · structuredmerge/structuredmerge-ruby

7.1.8 - 2026-09-07

TAG: v7.1.8
COVERAGE: 0.00% -- 0/0 lines in 0 files
BRANCH COVERAGE: 0.00% -- 0/0 branches in 0 files
73.88% documented

Added

Raise the minimum supported tree_sitter_language_pack version to 1.16.2,
the first released TSLP version containing the Ruby platform-gem metadata fix.
Add a repository-local benchmark report formatter that writes stable,
idempotent Markdown tables from JSON reports or standard input.
Add comprehensive --help output and usage examples to the repository-local
benchmark launchers.
Document all benchmark profiles and distinguish raw runs, aggregate reports,
and performance evidence in the benchmark guide.
Include benchmark selection mode, affected-path counts, neighbor counts, and
selected case IDs in Markdown reports so profile differences are visible.
Render configured competitor coverage and outcome summaries in competitive
Markdown reports, and document that competitor executables must be supplied
explicitly.
Expose the existing Git competitor adapter through --git so local
competitive runs can compare against the Git executable selected by PATH
without requiring the currently broken Mergiraf build.
Include competitor reported versions, executable paths, and binary digests in
the Markdown competitive report for reproducibility.
Make the repository-local benchmark launcher use the canonical reviewed sibling corpus by default while retaining explicit corpus selection.
Add a repository-local benchmark launcher that forces all cold workers to use the current StructuredMerge monorepo providers.
Preserve structured provider diagnostics in merge2 benchmark results so installed-stack and local-provider mismatches identify the actual unavailable provider.
Make YAML recursive resolution handle normalized MappingEntry nodes without assuming parser-wrapper predicates are present, restoring Psych provider conformance.
Make the local benchmark runner fail closed for non-Ruby corpus providers by recording them as unsupported coverage, support independently owned nested Markdown leaf sections while preserving ancestor framing, and compose independent direct-method edits inside Prism Ruby owners.
Add the generic ruby.ruby workflow provider over the Ruby substrate, accepting TreeHaver tree-sitter runtimes while keeping Prism as a distinct native backend provider.
Record the Ruby golden-master adapter identity, source revision and dirty state, executable digest, complete benchmark configuration, requested merge and parser providers, and allowlisted runtime environment in every local benchmark run and aggregate report.
Add a binary-safe, persistent JSONL benchmark adapter and performance-only run mode that measures many Ruby golden-master operations in one process without changing correctness classification.
Accept versioned adapter descriptors for cross-runtime benchmark runs, recording the candidate source state, executable digest, runtime environment, actual merge/parser providers, and declared support without mislabeling the candidate or treating out-of-scope neighbor cases as errors.
Add bounded process-level parallelism for independent cold benchmark cases with deterministic result ordering and explicit worker-process evidence.
Add explicit full-corpus nightly and competitor-enabled competitive benchmark profiles, and reject competitor sweeps from ordinary correctness profiles.
Separate benchmark process-spawn, first-request startup/loading, adapter execution, JSONL/IPC harness overhead, and warm round-trip timing evidence.
Add shared RSpec support for deterministic TreeHaver/provider contract snapshots, namespaced native-extension evidence, and serialization differential replay.
Add provider-owned Prism, Psych, RBS, and Markly snapshot extensions, and share one native RBS semantic projection between merge verification and contract evidence.
Add a clean-producer, one-process-per-replay capture driver for the seven representative Ruby golden-master provider snapshot targets.
Record workflow package identity and complete request/result transport evidence in provider snapshot differential replays.
Return unbalanced Ruby block directives as structured Prism provider parse failures instead of leaking exceptions.
Move source-preserving YAML provider mechanics into yaml-merge, add its TSLP workflow provider, and make psych-merge supply only its native parser projection.
Retain provider-owned native Psych AST extensions in production analysis results so coarse Alef workflow hosts can forward them without reparsing or flattening the native tree.
Let cross-runtime benchmark descriptors declare operation and dialect support plus merge-provider identity per format family, avoiding false support claims from Cartesian-product capability lists.
Load ast-merge-git's optional benchmark providers from STRUCTUREDMERGE_DEV during local development so benchmarks exercise the current monorepo providers.
Reuse YAML's recursive AST resolver in its workflow provider so nested template-only mapping leaves merge without sacrificing destination values or source-line provenance.
Route structurally rich TOML merge2 provider requests through the existing TreeHaver-backed TOML substrate merger, sharing table and dotted-key behavior across TSLP, Citrus, and Parslet providers.
Add the reviewed nested-table TOML merge2 sentinel to local benchmark profiles and recognize TOML provider identity in cross-runtime benchmark evidence.
Promote the shared Ruby class/method merge2 fixture into the local benchmark corpus and recognize the Rust Ruby substrate as a cross-runtime candidate.
Promote source-preserving Markdown heading-section merge2 into the mandatory benchmark profile and recognize the Rust Markdown substrate as a cross-runtime candidate.
Promote source-preserving TypeScript independent-function merge3 into the mandatory benchmark profile and map both Ruby and Rust TypeScript provider paths to affected-case selection.
Promote source-preserving Go independent-function merge3 into the mandatory benchmark profile and map both Ruby and Rust Go provider paths to affected-case selection.
Promote source-preserving Rust independent-function merge3 into the mandatory benchmark profile and map both Ruby and Rust-language provider paths to affected-case selection.
Promote source-preserving Bash independent-function merge3 into the mandatory benchmark profile and map both Ruby and Rust Bash provider paths to affected-case selection.
Add a reviewed Python merge3 gold case for the explicitly limited Rust
generic TSLP provider, including deterministic changed-path selection without
claiming a Ruby Python provider or adding it to the mandatory GM profile.
Validate the expanded reviewed-history corpus, including pinned Alef Rust
source and Markdown conflict resolutions, a clean JSON schema merge, and
reviewed Cargo TOML, Kubernetes Go/YAML, and RBS histories.
Record the Git baseline executable path, version, and digest in benchmark
environment provenance.
Add an opt-in competitive Git merge-file adapter with pinned corpus
provenance and the same strict version and digest checks as other tools.
Add per-competitor supported and eligible case counts with explicit coverage
ratios so comparative outcomes are not read without their denominators.
Add preservation-only line and lexeme agreement details to benchmark results;
these metrics do not affect merge correctness or safety gates.

Fixed

Align generated modular development Gemfiles with the nomono 1.1.5
bootstrap requirement used by the owning Gemfiles.
Let TreeHaver's TSLP adapter parse valid UTF-8 filesystem bytes whose Ruby
string is binary-labelled, without transcoding bytes or accepting invalid
UTF-8 as text.
Preserve destination-customized Markdown heading sections during provider merge2, insert parser-proven incoming-only sections according to shared ordering constraints, and reuse this source-preserving behavior across TSLP, Markly, Commonmarker, and Kramdown providers.
Let the RBS workflow provider honor explicit native RBS or TSLP TreeHaver selection, including the canonical kreuzberg-language-pack selector, across analysis, diff, and two- or three-way merge operations; route two-way operations through the RBS substrate so destination customizations survive; and reject TSLP's embedded inline-RBS parse mode for document merges.
Normalize object and hash tree-sitter points in shared line-range mechanics, and let RBS wrappers reuse that contract so TSLP merges do not fail on point shape.
Preserve shared layout-owned leading blank lines before template-only Prism Ruby nodes that have no leading comments.
Accept shared snake_case diagnostics from non-Ruby benchmark adapters so expected parser rejections are classified as successful fail-closed outcomes.
Compare YAML diff2 entries by semantic value and parser attributes rather than owned source fragments, so mapping order and formatting changes do not become semantic edits.
Normalize Citrus TOML pair ranges to their parsed value extent so comments following arrays are not emitted both as pair content and as shared owned comments.
Pin released StructuredMerge fixtures v7.1.8 for provider validation and benchmark profiles.
Run the Git benchmark competitor probe through a shell so CI can resolve the Git executable.
Make the default bup task commit CI-usable lockfiles against local StructuredMerge monorepo siblings while reserving deps-local for non-committing integration with unreleased Kettle sources.
Keep the root orchestration bundle on released tooling while passing local dependency overrides only to member bundle updates, preventing deps-local from polluting the root lockfile.
Run the default bupb task against local StructuredMerge siblings, matching the normal bup workflow.