Pietto

Readable, typed SQL authoring with deterministic compilation.

Pietto is a gradual semantic SQL authoring DSL. It parses and checks readable, indentation-based source, builds immutable compiler facts, and emits explicitly selected PostgreSQL or MySQL SQL. Version 0.1.4 adds a Python-style layout: clause items after : on the header line, separated by commas; condition blocks of one condition per line, joined with and; expressions that continue over several lines inside parentheses; and soft keywords such as type or select as names of fields, outputs, let: bindings and parameters. It also adds one-line relationships, relationship JOINs whose from line and VIA roles an explicit-module project infers when exactly one earlier binding and direction fit, and inline def functions whose calls emit the SQL of the same expression written by hand, and it fixes compiler and CLI defects found after 0.1.3. The alpha Python execution and saved-result API keep their explicitly selected database routes.

Version: 0.1.4 public alpha candidate. The versioned GitHub pre-release and PyPI listing determine release availability only after both channels provide the accepted files. Its compiler and CLI remain public surfaces; database access is available only from the explicit Python API, never from a Pietto CLI command.

What Pietto provides

• Typed shapes, sources, tables, queries, computed fields, aggregates, grouping, and bounded window expressions.
• Deterministic parser, semantic, and backend diagnostics with source locations where the relevant stage has them.
• Single-file check, explain, and emit-sql commands, with applicable JSON output, for tables and queries over one relation; JOINs, set operations, DISTINCT, relation lets, relation reads, query header parameters, typed VALUES bodies and generate_series need an explicit-module project.
• Explicit-project emit-sql for one selected table or query, one selected dialect and an explicit emission contract, producing the versioned pietto.sql-emission.v1 artifact (pietto.sql-emission.v2 for a VERIFIED query with header parameters).
• Deterministic project checking from an explicit pietto.toml root, including legacy-flat projects and the current explicit-module foundation.
• Compile-time validation and fail-closed lowering for unsupported syntax, semantics, and backend capability.
• Alpha Python execution through explicit PostgreSQL rows, MySQL rows, or PostgreSQL ADBC routes, with typed Arrow batches and factual outcomes.
• Alpha saved-result workflows: ordinary save, source-free R1 re-read, fixed-snapshot verified export, same-source R2 recovery, local publication, and relay to a local reference sink.

The compiler has no runtime evaluation, database connection, transaction management, scheduler, or arbitrary I/O language features.

Install

Install the published 0.1.4 files from PyPI after the release is available in a virtual environment with CPython 3.12 or later (replace python3.12 with your interpreter):

python3.12 -m venv pietto-env
pietto-env/bin/python -m pip install "pietto==0.1.4"
pietto-env/bin/pietto --version

The optional extras described below install the same way, for example pietto-env/bin/python -m pip install "pietto[arrow]".

After the release is available, its wheel and sdist are attached to the v0.1.4 pre-release. To install that wheel directly:

pietto-env/bin/python -m pip install "https://github.com/MianliWang/pietto/releases/download/v0.1.4/pietto-0.1.4-py3-none-any.whl"

When that release supplies SHA256SUMS, it lists the SHA-256 of the wheel and sdist. To check manually downloaded files, keep their names and run sha256sum --ignore-missing -c SHA256SUMS in the download directory. GitHub then also offers an automatic source archive under the same name, pietto-0.1.4.tar.gz; it is a snapshot of the v0.1.4 repository tag, not the sdist, and it is not listed in SHA256SUMS. The compiler and CLI were verified with CPython 3.12 and 3.13 on Linux x86-64; other platforms and Python versions are not covered by this evidence. With Pietto installed, run the commands below with pietto-env/bin/pietto (or activate the environment) instead of uv run pietto.

Version 0.1.4 rejects bundles and saved jobs produced by 0.1.3 or earlier by code identity; Pietto neither migrates nor rewrites those bytes, so retained state needs its matching old environment. The VS Code extension 0.1.0 accepts only the 0.1.2 wheel or its baseline fingerprint and refuses 0.1.4. Some programs that 0.1.3 accepted are refused, and some emit different SQL; see Compatibility and retained data.

Install from a checkout

Pietto requires Python 3.12 or later and uses a locked uv environment:

uv sync --locked
uv run pietto --version
uv run pietto --help

Optional Arrow interoperability

The core compiler and CLI need no PyArrow. The pietto[arrow] selector selects exactly pyarrow==26.0.0; the execution extras below add one database route's pinned drivers. From this checkout, install it into a separate environment:

uv venv /tmp/pietto-arrow
uv pip install --python /tmp/pietto-arrow/bin/python ".[arrow]"

Or build and select the extra on that exact local wheel:

uv build --wheel
uv pip install --python /tmp/pietto-arrow/bin/python "./dist/pietto-0.1.4-py3-none-any.whl[arrow]"

For a core-only wheel install, omit [arrow] and use a fresh environment. These are checkout/local-wheel instructions; from PyPI, install "pietto[arrow]" as shown above. The candidate package and CLI version are 0.1.4. The verified optional-extra matrix is CPython 3.12 and 3.13, Linux x86-64, PyArrow 26.0.0; other platforms, interpreters and Arrow versions are not covered by this evidence. Installing the extra only makes its dependency available; it does not select a route, grant execution, open a source, or qualify storage. The public result and execution APIs are alpha and described below. The Phase67 audit and handoff records the evidence layers, supported boundaries and remaining execution obligations.

Execution extras

Each execution route has its own selector; each includes the Arrow pin because captured results are Arrow IPC:

Saved-result use without any database driver needs only pietto[arrow]. Selectors combine, for example "pietto[execute-postgres,execute-mysql]". Installing an extra never selects a route, grants execution, opens a source or qualifies storage. A selected route whose pinned driver is absent refuses with EXECUTION_DEPENDENCY_MISSING before any connection, another version refuses with that route's version error, and another route's installed driver is never used instead. These extras were verified on CPython 3.13.13, Linux x86-64, with the lock's exact artifacts; other platforms, interpreters and driver versions are not covered. The execution alpha below is part of the 0.1.4 candidate; it becomes available from PyPI and GitHub assets only after the accepted files appear on those channels.

Execution alpha

pietto.execution is an alpha Python interface for explicit, bounded database attempts and saved results. It compiles a project query or loads a query bundle under caller-held trust records, then returns typed Arrow batches and the attempt's factual terminal state. It never accepts SQL text and Pietto has no CLI command that connects to a database.

The direct routes are postgres_rows, mysql_rows and postgres_adbc; each requires its selected extra, explicit access and a deployment premise. The saved-result workspace provides ordinary save, R1 source-free re-read, fixed-snapshot verified export, R2 same-source recovery, local publication and relay to a local reference cooperative sink. R2 takes one caller-supplied compiled-source declaration for every source in compiled order, including guarded and refined sources. The API reference records effects, cleanup and compatibility requirements in the execution documentation.

All execution names, fields, PIE-R codes and behavior may change in a later alpha release. The compiler, CLI and existing public contracts do not change. The native routes have evidence only on CPython 3.13.13, Linux x86-64, with the lock's exact artifacts on the qualified integration host. Windows native execution is not verified.

Quick start

Create active_users.pietto:

shape User:
    id: Int not null
    email: Text nullable
    email_norm: Text nullable
    deleted: Bool not null

source users: User is postgres.table("users")

table active_users:
    from users
    where deleted == false
    select:
        id
        email
        email_norm = lower(trim(email))

Check it:

uv run pietto check active_users.pietto
uv run pietto check active_users.pietto --format json

Explain its semantic metadata:

uv run pietto explain active_users.pietto
uv run pietto explain active_users.pietto --format json

Generate PostgreSQL SQL:

uv run pietto emit-sql active_users.pietto --dialect postgres

check does not generate or execute SQL, so a file that passes check can still be refused by explain and emit-sql (an authored qualify: block is PIE-I1000) or by the selected dialect (ignore nulls is PIE-B1000); only emit-sql --dialect X shows that a file emits for X. emit-sql writes SQL to stdout by default and can atomically replace an explicitly selected regular output file. Text output lists the SQL bodies in backend order separated by one blank line, without artifact names or ;; use --format json to see which artifact is which. When there is no SQL to write, for example for a file without a table or query, emit-sql --output exits 2 and leaves the target file untouched. pietto with no arguments prints the top-level help to stdout and exits 0. Terminal text uses the terminal's selected encoding; if it cannot represent an output, Pietto exits 2 with an ASCII-safe error instead of replacing or partially writing SQL. SQL files selected with --output remain UTF-8 and use LF line endings on every platform. check, explain, and emit-sql take --check-mode loose, checked (the default), or strict for one file. The language guide describes the modes.

A single-file postgres.table or mysql.table string may be one table name or <schema>.<table> (for MySQL, <database>.<table>); each part is quoted separately, so "public.users" emits FROM "public"."users", and more than one dot or an empty part fails check with PIE-S2306. Project SQL emission takes physical schema and relation names from its emission contract.

Project SQL emission carries Timestamp and UUID fields only as the contract-qualified physical values of its emission contract: Timestamp requires PostgreSQL timestamp without time zone(6) or MySQL DATETIME(6) and the fixed civil law, and UUID requires PostgreSQL uuid or MySQL BINARY(16) and standard big-endian bytes. On those carriers it admits carrying and projecting the values, ==/!=, count, count_distinct, DISTINCT, GROUP BY, and direct inner/left/right equality joins; ordering and inequality comparison, min/max, full joins, a Timestamp or UUID window partition or order key, range/as-of behavior, timezone conversion, parsing and other casts remain unavailable. See Timestamp and UUID source meaning.

Projects

Project commands use an explicit root containing pietto.toml:

demo-project/
├── pietto.toml
└── models/
    └── active_users.pietto

For the available legacy-flat project mode:

schema_version = 1

[sources]
include = ["models/*.pietto"]

Then run:

uv run pietto check --project demo-project

Project input selection and diagnostics are deterministic. Project checking is not SQL generation; project SQL emission is a separate explicit command.

Project SQL emission

emit-sql --project compiles one explicitly selected table or query of an explicit-module project into the versioned pietto.sql-emission.v1 artifact (pietto.sql-emission.v2 for a VERIFIED query with header parameters, see project CLI JSON v2) for one explicitly selected dialect. The physical relation and column names, storage types and value premises come from an emission contract you write for your database; the compiler never connects to it.

demo-emit/
├── pietto.toml
├── main.pietto
└── contract.json

pietto.toml selects the explicit-module project mode:

schema_version = 2

[sources]
include = ["*.pietto"]

main.pietto:

shape Item:
    id: Int not null
    label: Text not null

source items: Item is postgres.table("fixture.items")

table active_items:
    from items
    select:
        id
        label

contract.json maps the logical source items to the physical relation fixture.items and each field to its column and representation:

{
  "format": "pietto.emission-contract.v1",
  "target": {"family": "postgres", "release": "18.6"},
  "sources": [{
    "selector": {"module": "main.pietto", "kind": "source", "name": "items"},
    "relation": {"namespace": "fixture", "name": "items"},
    "scan": "relation_rows",
    "fields": [
      {"ordinal": 0, "name": "id", "column": "item_id",
       "representation": {"storage": {"kind": "pg_int8"}, "nullable": false,
                          "domain": {"kind": "int_range", "min": "0", "max": "100"}}},
      {"ordinal": 1, "name": "label", "column": "item_label",
       "representation": {"storage": {"kind": "pg_text"}, "nullable": false,
                          "domain": {"kind": "text", "max_characters": 64,
                                     "encoding": "UTF8", "collation": "C",
                                     "padding": "NO PAD"}}}
    ],
    "premises": [
      {"key": "row_domain_matches", "scope": "source", "value": true},
      {"key": "read_only_object", "scope": "source", "value": true}
    ]
  }],
  "environment": [
    {"key": "client_encoding", "scope": "statement", "value": "UTF8"},
    {"key": "operator_environment", "scope": "statement", "value": "builtin_only"}
  ]
}

This table compiles to one plain SELECT. On PostgreSQL, a table or query whose SQL needs named stages (a WITH clause), for example one with where, group by, order by, a window, a JOIN, a relation let or a set operation, also needs the statement premise {"key": "identifier_case", "scope": "statement", "value": "quoted_exact"} in environment; without it the artifact is BLOCKED with PIE-B1004 (identifier_case_declaration_missing).

Emit the artifact as JSON on stdout (--format json and --literal-policy preserve are the defaults; the contract path is relative to the project root):

uv run pietto emit-sql --project demo-emit --module main.pietto --kind table \
    --name active_items --dialect postgres --emission-contract contract.json

Show the readable presentation on stdout and atomically write the same JSON artifact to a file:

uv run pietto emit-sql --project demo-emit --module main.pietto --kind table \
    --name active_items --dialect postgres --emission-contract contract.json \
    --format text --output active_items.sql-emission.json

Exit codes are 0 for a VERIFIED artifact, 1 for BLOCKED (typed blockers or project diagnostics) and 2 for INPUT_REJECTED (usage, project, contract or output errors); failures carry no candidate SQL. A --literal-policy bind-safe artifact must be consumed together with its fixed_values and parameter_uses; its SQL alone is not a complete statement. The complete contract is the Slice13 specification.

Documentation

• Language
• Project and package
• Development
• Roadmap
• Status

Public contracts:

• CLI JSON v1
• Diagnostics
• Project JSON v2
• Semantic metadata artifact v1
• Configuration
• Golden fixture policy

Development

Run the authoritative local validation suite with:

uv run python scripts/validate.py

The normal implementation loop is focused tests, Ruff, and targeted type checking. Natural CI runs the final Python 3.12 and 3.13 validation coverage.

Direction

Pietto is an early preview. These are the goals it is working toward; they are not promises or dates, and the detailed plan lives in the roadmap.

• Everyday SQL: CASE, CAST, COALESCE/NULLIF, IN lists, NOT, LIKE, and common string, number and date functions.
• Real-world column types end to end: timestamps, UUIDs, decimals, dates and times, then arrays and nested records.
• Richer queries: subqueries and WITH, EXISTS/IN, VALUES and table functions.
• Richer aggregation: SUM and AVG in project SQL emission, FILTER, statistical aggregates, grouping extensions and aggregates as window functions.
• Run queries safely from Python: compile, execute through an explicit database driver, and get typed Arrow results.
• Data tools: pandas, Polars and NumPy interoperability.
• More databases: deeper PostgreSQL and MySQL support, then SQLite and DuckDB.
• Tooling: a formatter, language-server and editor support, clearer diagnostics.
• Stability: a 1.0 release with frozen syntax, CLI and APIs.

License

Pietto is released under the MIT License. Third-party material in the wheel, the sdist and the source archive is listed in THIRD_PARTY_NOTICES.md, with the original license texts in LICENSES/. The Pietto logo artwork in docs/assets/brand/ is © Mianli Wang, all rights reserved, and is not covered by the MIT License that applies to the code.