Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[Unreleased]

Write-pass atomicity for codegen: prepare every sql.ex write before committing; refuse all writes when any prepare fails; commit via temp + rename with rollback if a later rename fails (#49).

Directory discovery returns structured CannotReadFile instead of raising on unreadable paths.
Removed unused internal Codegen.write_directories / check_directories helpers (partial progress on #52).

Clarify query-error vs write-pass atomicity in README, Configuration, Phoenix guide, and Mix task docs. ROADMAP points at Post-0.5 backlog and 1.0 issues.

Phoenix-style lib/my_app/.../sql paths generate MyApp....SQL (no duplicated app segment).
Bang/? query names sanitize @type/@spec row identifiers; soft companions deconflict against query names and other soft companions with a warning.
Connection URLs percent-decode userinfo, reject schemeless input, and no longer let URL defaults clobber present PG* values for omitted fields.
mix precommit compiles with --warnings-as-errors; Hex publish runs credo + tests before publishing.
EXPLAIN DO-block warning is captured in tests; inference errors with code: nil no longer format as [NIL].

Multi-schema / search_path inference docs (Configuration + Writing Queries) and ExUnit coverage for unqualified tables resolved via session search_path and schema-qualified tables outside search_path.
CI test coverage via ExCoveralls: HTML artifact + GitHub Actions job summary, soft floor warning (not a hard fail). Local mix test / mix precommit unchanged; opt in with mix cover / mix coveralls.html.
Adopter CI workflow examples under examples/github-actions/ for live mix squirrelix.check --infer and offline metadata check.

Generated query modules are Dialyzer-friendlier under typical adopter flags: encode helpers gain matching @specs and guards, command num_rows helpers document non_neg_integer(), UUID helpers use precise binary specs, and underspec'd private decode_rows/decode_row contracts are omitted. Types and Phoenix guides document Dialyzer expectations and known :overspecs limitations (no Dialyxir dependency).
User-facing errors for file I/O, missing/incomplete/invalid metadata, invalid connection URLs, and invalid CLI options use the same structured titles/hints as connection and query diagnostics (no raw inspect/1 primary messages). Existing structured connection error types from 0.2/0.3 are unchanged.
Documented and narrowed the supported pre-1.0 public API: Squirrelix.generate/3 / check/3, Mix tasks, codegen summaries, documented Squirrelix.Error.* structs / Error.format, Postgres.inferrer/1, Inference.Inferrer, and Query (Inferrer argument). Other library modules are marked @moduledoc false / internal.

mix squirrelix.gen --watch watches {lib,test,dev}/**/sql/*.sql and regenerates on change (same query source / connection options as a one-shot gen; Ctrl-C to stop). Linux needs inotify-tools.
--write-metadata PATH on mix squirrelix.gen / mix squirrelix.check (requires --infer) exports a reloadable metadata file for offline check/codegen without Postgres.
Broader structural parameter naming beyond INSERT: comparison operators (<>, !=, <, >, <=, >=), LIKE/ILIKE (including NOT), and SET (columns) = (...) / ROW(...) lists (including ON CONFLICT ... DO UPDATE SET). Comparison/equality names still win over column-list inference; existing UPDATE ... SET col = $n and INSERT (cols) VALUES naming are unchanged.
Phoenix + CI cookbook guide (guides/phoenix.md) for Mix adoption: migrate-then-gen/check, DATABASE_URL, Mix aliases, CI check jobs, and intentional Ecto coexistence.

Generated soft companions (<name>_ok/arity) via Postgrex.query/3 that return {:ok, result} | {:error, Exception.t()} without raising. Soft command companions return {:ok, num_rows} (affected-row count). The raising query! API is unchanged (additive, not breaking). See Writing Queries and Getting Started guides.
Parameter name inference from INSERT ... (columns) VALUES (...) placeholders (equality / UPDATE ... SET naming unchanged; equality wins on conflicts).
--infer honors DATABASE_URL with documented precedence: flags → --url → DATABASE_URL → PG* → defaults. URL/PGSSLMODE sslmode (and ssl=true / ssl=false) map into Postgrex :ssl options.

Project-wide atomic codegen (Gleam squirrel 4.5+ parity): if any sql/ directory has query errors, mix squirrelix.gen / Squirrelix.generate/3 write nothing. mix squirrelix.check fails globally when any directory has errors or drift.

Nullability inference for schema-qualified tables, scalar subqueries in select lists, and expression-derived columns (see Writing Queries guide).
Structured errors for --infer connection failures and timeouts (CannotConnectToPostgres, PostgresConnectionTimeout) with actionable hints for PG* env vars, CLI flags, and metadata-file fallback.

Documented composite-type policy as reject-with-hints (aligned with Gleam Squirrel and flat Elixir row maps). Inference now attaches actionable hints for composites (kind: "c") and geometric point values.
Finalized unsupported-type policy for Postgres ranges/multiranges and remaining built-ins (interval, geometric, network, money, and related types): inference rejects them with actionable hints, and the Types guide documents workarounds.

First public release of Squirrelix (Hex package squirr_elix): typed Elixir query modules from plain .sql files via Postgres inference or metadata.
Mix tasks mix squirrelix.gen and mix squirrelix.check.
Guides for getting started, writing queries, types, and configuration.
GitHub Actions CI (format, compile, Credo, tests with Postgres 16) and Hex publish on v* tags.

Public Elixir modules and Mix tasks use the Squirrelix name (package/app remain squirr_elix). Repository: scripthungry/squirrelix.

Generated SQL and @doc strings are embedded with inspect/2 so #{} in .sql files cannot become Elixir interpolation in generated modules.