dabiprobe

Check your D bindings against what the C compiler actually sees!

dabiprobe helps maintainers catch binding regressions when declarations, compilers, or build flags change. It compiles independent C probes, reflects the selected D declarations, and runs optional cross-language assertions. It does not generate bindings or replace ImportC.

Build and run

Requires a GCC/Clang-compatible C11 driver and LDC or DMD targeting the same runnable host ABI. The verifier uses only the D standard library.

mkdir -p .artifacts
ldc2 -i -Isource -of=.artifacts/dabiprobe -od=.artifacts source/app.d

./.artifacts/dabiprobe --output .artifacts/packet-check examples/packet/manifest.json

Alternatively, dub build writes the executable to .artifacts/. Windows executables use .exe. --cc and --dc select compiler executables; defaults are cc (clang on Windows) and ldc2.

Reuse your declarations

Select names rather than maintaining another copy of the binding:

{
  "headers": ["packet.h"],
  "imports": ["packet"],
  "types": [
    {"c": "struct Packet", "d": "Packet", "fields": "*"}
  ],
  "constants": [
    {"c": "PACKET_LIMIT", "d": "packetLimit"}
  ]
}

fields: "*" discovers the D members. Function entries can likewise contain only c and d names; the D compiler supplies their signatures. Checks include field representations, array elements, and supported callback types—not just sizes. Explicit field mappings and function contracts remain available. Opaque handle mappings use opaque: true; their object layout stays unchecked.

Supply the project's actual compiler arguments in cFlags and dFlags. Relative input paths resolve from the manifest directory. Named configurations append variant-specific flags; run all variants or select one with --config NAME.

The BindBC-GLFW case study uses a pinned real binding, headless library calls, and clearly labeled injected regressions. The smaller packet and interop examples introduce the basic checks.

Read and reproduce results

Results distinguish measured agreement from executed behavior assertions. Each check identifies its declaration, property, configuration, and outcome:

Outcome	Meaning
checked	Selected measurement agrees, or an explicit assertion returned true
mismatched	A measured property or assertion disagrees
unchecked	Property unexamined, or execution deliberately skipped
unsupported	A requested contract cannot be represented by the checker
failed_execution	A required operation could not complete

Exit codes are 0 for no disagreement in supported executed checks, 1 for a mismatch, 2 for an input/execution error, and 3 for an unsupported or otherwise incomplete requested check. Unchecked and unsupported properties never count as passed checks.

--output NEW_DIR saves report.json, the effective configurations, generated sources, exact commands, and diagnostics. Probe failures otherwise retain private evidence and print its location. Existing output directories are never overwritten.

python3 .artifacts/packet-check/replay.py

Replay rebuilds the captured probes without dabiprobe. Headers, binding modules, libraries, and compatible compilers remain external dependencies. Changed D declarations or targets require a fresh dabiprobe run, not stale generated checks.

Scope

A pass is evidence for the selected properties and assertions, not the entire binding. C-member completeness, ownership, lifetimes, and untested behavior are not inferred. Header agreement does not certify an arbitrary prebuilt library.

Use trusted inputs. Compilation, constructors, and assertions can execute code; there is no sandbox or subprocess timeout. Evidence may contain project sources, paths, and build flags, so review it before sharing.

See dabiprobe --help, the manpage, and the related-work comparison for details.

Test

python3 tests/integration.py .artifacts/dabiprobe --cc cc --dc ldc2

The suite uses real compilers. Python 3.10+ is required; Windows replay tests need Python 3.13+ for private temporary-directory support.

License

BSD 2-Clause. Copyright (c) 2026 GuestAUser.