Getting started

This guide takes an existing CMake project from nothing to a verified ReproBit build. Every command below was run in that order against a fresh copy of the examples/grind-progress sources; the printed output is reproduced with only the absolute project path shortened to /path/to/project. For the shortest possible first contact, run the ready-made grind example instead; for what the checks mean, read Concepts. This page owns the normal setup sequence. The command-line guide covers optional and advanced behavior without interrupting that path.

Prerequisites

• Python 3.11 or newer with ReproBit installed (README).
• The project directory is a Git working tree with its source files committed. rbit source reads the Git-tracked file list, so untracked files are invisible to it.
• The project's compatible CMake version on PATH (used only by CMake import and later source-list refreshes).
• On macOS and Linux, Wine and wineserver on PATH.
• If the first build will need automatic grind, project-owned reference .obj files for the source files you want it to search. The reference executable alone is enough to verify, but it cannot supply function bodies to grind. You can skip these objects when the project already matches.
• A short project path. MSVC 4.2 has small internal path buffers: a first attempt under a 150-character path failed inside CMake's compiler test with fatal error C1005:; the same project under a 55-character path worked. See Troubleshooting.

The sequence

rbit init . --target program
rbit setup .
# Add /reference/ and /build/ to .gitignore before reviewing source files.
rbit source preview .
rbit source lock .
# Put the reference binary at reference/program.exe, then:
rbit import cmake .
rbit status .

rbit status . works at every point in between and prints the next command, so you never have to remember where you stopped. Each step is explained below.

1. Create the project entry point

Name the real CMake target when initializing. Repeat --target for projects that produce more than one binary; the default rebuilt-output and reference filenames follow each target name (build/NAME.exe, reference/NAME.exe). When the CMake output name differs from the target name, declare both paths:

$ rbit init . --target grind_progress --artifact build/grind-progress.exe --oracle reference/grind-progress.exe
Created ReproBit project 'gs' at /path/to/project
Next: rbit setup /path/to/project

With several targets, qualify custom paths as --artifact TARGET=PATH or --oracle TARGET=PATH. Use --target program=your_cmake_target during import cmake only when the ReproBit target ID must intentionally differ from the CMake target name and its declared artifact still matches the real output.

init writes reprobit.toml, adds root .gitignore entries for ReproBit's local state, and creates an empty, explicitly incomplete source record, so a new project cannot appear build-ready before its real files are reviewed. It does not touch CMakeLists.txt. status now lists everything that is still missing:

$ rbit status .
Project: /path/to/project
Project and machine: 3/12 checks ready
[  ] Compiler lock: run rbit setup to create reprobit/toolchain.lock.json
[  ] Source lock: run rbit source preview to finish reviewing and locking the tracked files
[  ] Protected references: place the original at reference/grind-progress.exe
[  ] Build plan: run rbit import cmake to create reprobit/build-plan.json
[  ] Build graph: run rbit import cmake to create reprobit/producer-graph.json
[  ] Interventions: run rbit import cmake to create reprobit/interventions/
[  ] Proofs: run rbit import cmake to create reprobit/proofs/
[  ] Reference metadata: run rbit import cmake to create reprobit/oracles/
Next: rbit setup /path/to/project

status exits with 1 while anything is missing; that is an honest "not yet", not an error (see Exit status).

2. Prepare this machine

$ rbit setup .
authenticating the compiler installation...
authenticating the compiler installation: complete (0.0s elapsed)
locking exact compiler files...
locking exact compiler files: complete (0.0s elapsed)
Environment ready for Microsoft Visual C++ 4.2
Compiler: /Users/you/Library/Application Support/ReproBit/toolchains/msvc_4_2
Project lock: created
Project: /path/to/project
Project and machine: 4/12 checks ready
[  ] Source lock: run rbit source preview to finish reviewing and locking the tracked files
[  ] Protected references: place the original at reference/grind-progress.exe
...
Next: rbit source preview /path/to/project

setup downloads and authenticates the compiler when this machine does not have it yet (the first run therefore prints a download activity as well), creates reprobit/toolchain.lock.json when it is absent or checks an existing lock against the installation, and probes the host backend. After those checks pass, it remembers the compiler location in the user configuration. A failed setup preserves the previous default. Later commands need no machine-specific paths. Pass --toolchain-root /path/to/msvc42 to use an installation you already have; platform setup and the native Windows guide cover unusual hosts.

Machine setup and project setup are separate: setup can finish while status still lists missing project files, and successful compiler setup alone never counts as a build-ready project. Run setup once on each machine that will build the project.

3. Review and lock the source files

Before locking, add /reference/ and /build/ to .gitignore unless you intend to commit those directories. init adds ignores for ReproBit's local state; these project output and reference directories are your choice. Finish these edits now: tracked .gitignore files belong to the source read set, so editing one after source lock requires reviewing and locking it again.

$ rbit source preview .
checking the project source files...
checking the project source files: complete (0.0s elapsed)
Source preview: +4 -0 ~0; 4 selected inputs
  add: CMakeLists.txt, prepare_reference.py, transform_one.cpp, transform_two.cpp
  no build plan or saved source-derived records to check
Next: rbit source lock /path/to/project

$ rbit source lock .
checking the project source files...
checking the project source files: complete (0.0s elapsed)
locked 4 project source inputs
Next: place the original at reference/grind-progress.exe

source preview hashes the proposed Git-tracked read set without writing; source lock publishes that set to reprobit/source-manifest.json after your review. Repeat --path on either command to provide an explicit complete file or tree set instead. It is not a filter over the current record: every omitted path is reported as a removal.

4. Place the reference inputs

Copy the original executable to the declared oracle path (reference/grind-progress.exe here). ReproBit never redistributes reference binaries. The ignores added before source review keep these inputs and build outputs out of Git (the shipped examples do the same).

If you expect to use automatic grind after the first verification, also place the available reference objects under reference/. Name each one after its source filename without the extension—for example, src/widget.cpp pairs with reference/widget.obj. Grind's read-only preflight reports how many eligible compiler steps have an object, how many do not, and how many functions it selected. Use --reference-object TU=PATH when names are ambiguous. The examples/grind-progress script generates both kinds of reference input with the authenticated compiler; a real project uses its own archival or analysis inputs. The grind guide covers the full mapping rules.

5. Import the build steps from CMake

$ rbit import cmake .
preparing a reviewable CMake build plan...
preparing a reviewable CMake build plan: complete (0.1s elapsed)
configuring the CMake project...
configuring the CMake project... (5.0s elapsed)
configuring the CMake project: complete (52.0s elapsed)
recording the direct compiler and linker steps...
recording the direct compiler and linker steps: complete (0.0s elapsed)
committed 3 direct producers to reprobit/producer-graph.json
CMake import complete: 3 build steps and 2 TUs recorded
Next: rbit build /path/to/project

import cmake derives the initial build plan, records the reference binary's digest, and creates an empty review document for each unambiguous source file so discovery can start without hand-authored JSON. It configures the existing CMake project without building it and saves the direct compiler and linker steps as reprobit/producer-graph.json. It derives only facts it can check and does not guess entropy interventions. It does not edit CMakeLists.txt, and normal ReproBit builds never invoke CMake. If the import fails, the generated scaffold is removed and the diagnostic workspace is retained for inspection. CMake import and refresh covers projects that need cache settings, later source-list changes, or the configure and extract halves separately.

6. Confirm and commit

$ rbit status .
Project: /path/to/project
Project and machine ready: 12/12 checks passed

$ git status --short
?? .gitignore
?? reprobit.toml
?? reprobit/

Commit .gitignore, reprobit.toml, and reprobit/ (the compiler lock, the source lock, the build plan, the build graph, and one empty intervention and proof document per target and per source file). rbit validate . loads every saved JSON file, rejects duplicate keys and IDs, checks cross-document references and dependency cycles, compares current files with their recorded hashes, and never runs a build:

$ rbit validate .
validated gs: 1 target, 0 interventions

7. Build, then verify

$ rbit build .
checking the project files...
rebuilding changed steps and reusing unchanged work: complete (8/8; 7.8s elapsed)
Incremental build: 0 reused, 6 rebuilt (0.0% reused); compiler environment started 1 time; 7.77s; target outputs: 0 unchanged, 1 updated
Why steps were rebuilt:
  compiler.grind_progress.0000: not cached on this machine yet (first build of this step)
  compiler.grind_progress.0001: not cached on this machine yet (first build of this step)
Build complete: 6 steps, 1 output
  build/grind-progress.exe (1,536 bytes)

build is the incremental developer loop. Running it again without any edit restores every step from the cache without starting the compiler environment:

Incremental build: 6 reused, 0 rebuilt (100.0% reused); compiler environment started 0 times; 0.30s; target outputs: 1 unchanged, 0 updated

Use rbit build . --cold for a non-certifying build with no cache reads or writes. Certification is a different command:

$ rbit verify . --report-dir build/reprobit-report

verify always builds from scratch, never reads the developer cache, compares the result with the protected reference, and writes report.json plus a self-contained report.html. Open the HTML in a browser: its first view explains the overall result and each target; detailed symbols, commands, and evidence are in collapsed Advanced sections. A freshly imported project usually does not match yet. This one prints:

Verification did not satisfy the authenticity policy
Byte identity: 0/1 targets exact
Report: /path/to/project/build/reprobit-report/report.html

and exits with 1. That is the starting point for the bounded discover grind workflow, which finds the first low-cost adjustments and saves them only after a fresh exact proof. Its first line checks the reference-object coverage before compiling. Once the project is exact, verify ends with Verification passed, Authenticity: clean, and the total intervention cost.

Debug companions and the exported source view

The declared binary remains the only output used for byte-exact certification or release. If an imported MSVC link asks for debug data, ReproBit also writes a matched binary and .PDB under the sibling reprobit-debug/ directory—for example, build/reprobit-debug/GAME.EXE and build/reprobit-debug/GAME.PDB. Give analysis tools those two files together; do not mix the .PDB with the declared binary. ReproBit chooses the paths automatically and caches the pair during incremental builds.

If the project uses reviewed source adjustments, export the matching source view before running a source-aware comparison tool, and point the tool's source root at that directory:

rbit source export . --destination build/reprobit-debug/source

This keeps line and symbol information matched to the files the compiler actually read. The export contains project inputs admitted by the source lock, with the reviewed source adjustments applied, plus one hidden ownership marker; it does not contain reference binaries, compiler files, build outputs, or private run state. Run the same command after later changes; ReproBit safely replaces the previous view and removes files that are no longer part of it.

After the project is exact: edits and repair

After editing a file in a project that already matched exactly—even a shared header used by many source files—let ReproBit repair and verify the project in one step:

rbit repair .

repair works on a private copy, follows shared headers to every affected source file, refreshes the saved guidance it can prove, and checks every target from scratch. It may take several passes, but publishes the records, binaries, debug companions, and report together only after the whole project is exact and trustworthy. If it cannot prove the result, your source edit remains while previously published records and results stay unchanged. No hand-written JSON or TOML repair recipe is needed.

This is the maintenance path once a project has reached an exact match; a project that builds but does not match yet starts with discover grind instead. See rbit repair for report paths and advanced search budgets.

To add a new file to the reviewed source list—or remove a locked one—start with rbit source preview .; repair never silently admits newly tracked files. Before the first import, preview prints a safe source lock command. After an import, it normally prints rbit import cmake . --refresh instead. That command prepares the new source list and CMake records privately, verifies every target from scratch, and publishes the records, verified outputs, and report only as a complete passing update. If it cannot keep the existing reviewed records safely, it refuses without changing the published project. See Refreshing a CMake import.

Reclaiming space

Failed builds keep a private workspace so problems can be inspected. Reclaim those workspaces with rbit clean .; the reusable build cache is kept by default. rbit state status . points out cache data the current ReproBit version cannot reuse. Include it when cleaning inactive workspaces with rbit clean . --obsolete-cache --preview; the current cache is kept. Use --cache to clear the complete cache, or --reports to remove generated verification and grind reports.

CI and machine-readable output

ReproBit uses the remembered authenticated compiler and the locked host launchers. CI can still pass explicit machine paths (--toolchain-root, --compiler-transport/--resource-transport) and emit NDJSON with --format ndjson; the GitHub Action packages the build-from-scratch verification workflow. The event format is documented under Machine-readable output.