FrostBuild Documentation

Build with the full picture.

Start with a working build, then go as deep as you need. The architecture and manifest specification are the contract; every performance claim links back to a reproducible report.

Start here ↓ Browse Markdown

v0.8.0 · quick start

cargo install --git \
https://github.com/hjosugi/frost-build \
--tag v0.8.0 frostbuild-cli

frost init
frost build

Native C/C++ and plain Java can be scaffolded automatically. Other ecosystems use explicit, direct-command boundaries.

Source of truth DESIGN.md defines the engine. The manifest specification defines frost.toml. Research documents explain why.

START HERE

Three documents.
One mental model.

Learn the commands, the correctness contract, and the configuration format before exploring implementation studies.

01 · Guide

Quick start

Install FrostBuild, initialize a workspace, build, test, run, and inspect a graph.

Open README → 02 · Normative

Architecture

The action model, graph, cache boundaries, invariants, non-goals, and v2 line.

Read DESIGN.md → 03 · Reference

Manifest specification

Every supported frost.toml field, target type, profile, and platform rule.

Open the specification →

BUILD AND INTEGRATE

From one target
to a polyglot graph.

Use native rules where Frost owns the compiler boundary and explicit adapters where an ecosystem already owns its inner build.

Everyday workflows

Build, test, watch, debug, and migrate existing graphs.

Developer loop Host and target platforms Conservative Bazel migration Ninja importer

Language and ecosystem boundaries

Choose the unit Frost can cache and verify without hiding work.

Language adapter contract Polyglot win matrix npm workspace import npm production adoption

PERFORMANCE EVIDENCE

Read the scope,
then the number.

Reports preserve equivalent work, semantic artifact checks, raw samples, tool versions, and losing cases.

Method and system comparisons

How results are collected and where FrostBuild still loses.

Benchmark methodology Bazel gap analysis Build-tool landscape Raw benchmark baselines

Equal-artifact language reports

One compiler and artifact contract at a time.

Java · Gradle · Maven Rust · Cargo Go · go build TypeScript · tsc Python wheels

ARCHITECTURE AND RESEARCH

Decisions keep
their evidence.

These documents explain the trade-offs behind pruning, storage, scheduling, remote work, and migration.

Core strategy

The mechanisms behind correctness-first incremental work.

Nix, Bazel, and micro-partitions 2× strategy Cache-layer separation Action-key audit

Forward-looking studies

Measured options, explicit adoption decisions, and deferred work.

Papers and primary sources Remote cache Remote execution Predictive selection Learned scheduling DeltaCDC calibration

PLATFORMS AND PROCESS

Know what is tested.
Know what is historical.

CI coverage, fuzz targets, issue-to-implementation evidence, and superseded design paths stay visible.

Correctness and delivery

Verification surfaces and their tracked implementation evidence.

Fuzzing and property tests Issue implementation matrix Contributing

Historical record

Useful context that is no longer the product contract.

Superseded Zig plan July 2026 session log Complete documentation index

Can’t find the boundary you need?

Open an issue with the language, artifact, incumbent build tool, and a reproducible clean/no-op/change workload.

Open an issue