Tools

Helper programs for working on badc. None of them is part of the compiler or needed to build it.

tools/

path	what it does
gen_isa.py	generates the x86-64 encoder table src/c5/codegen/x86_64/isa_x86_table.rs from asmjit’s isa_x86.json
gen_isa_a64.py	generates the AArch64 encoder table src/c5/codegen/aarch64/isa_a64_table.rs from asmjit’s isa_aarch64.json
gen_sysreg_a64.py	generates the AArch64 system-register and system-operation tables by disassembling every selector with a reference assembler
scaling_probe.py	measures how compile time grows with input size, one generated input shape at a time
core-walker.py	names the frames of a Linux x86-64 core dump when a debugger cannot
diff_vorbis/	traces stb_vorbis.c built by clang and by badc and diffs the traces, to find the first function whose behavior differs
probe_asm_units/	passes each preprocessed .S unit through badc’s file-scope asm path and ranks the first diagnostics

The generated tables are checked in, so the generators run only when a table changes.

core-walker.py walks the saved-rbp chain of a core dump and reports each saved return address as a file offset in the original non-PIE binary, whose load base it takes as 0x400000:

• with no option, it walks the rbp chain;
• --dump-around-rbp prints the sixteen 8-byte slots around rbp;
• --scan-stack ignores the chain and resolves every 8-byte slot above rsp that looks like a code address, which finds return addresses when stack corruption has broken the chain;
• --list-segments lists each PT_LOAD segment with its address range.

probe_asm_units/ measures the assembler alone, with no code model and no dependency output. The kernel page has the build’s own figures.

scripts/

The scripts used most:

script	what it does
cargo_test.py	runs cargo test and also fails a test binary that stops before reporting every test
snapshots.py	regenerates tests/snapshots/; --check fails on any difference
coverage.py	line and function coverage of src/ under the test suite
csmith_fuzz.py, c_reduce.py	differential fuzzing with csmith, and reduction of a failing case
run_demos.py	runs demo smoke tests concurrently
compile_throughput.py	compile-time ratios over a fixed corpus, checked against ceilings
diagnostics_doc.py	writes diagnostics from the diagnostic catalogue
install_hooks.py, validate_local_boxes.py	the git hooks and the pre-push gate
perf_publish.py	files a benchmark run in the data repository the performance page reads

Testing describes how they fit together.

^ To the top