codex-cup-release

---

name: codex-cup-release description: Use when Codex needs to build, test, or modify an external C/C++ project that uses only a Cup release artifact: build/embedded/cup.exe or build/header_only/cup.h. Covers first-time setup, embedded vs header-only detection, cup -hh, cached options, target paths, tests, ENTRY DAG editing, and correct macro recipes including LIB+AR for static libraries and EXE/DLL+LINK for linked binaries.

Cup Release Usage

Use this skill in a consumer C/C++ project that uses Cup as its build system. Do not assume Cup source files exist. The user may only have one downloaded release artifact:

• build/embedded/cup.exe on Windows, or an equivalent executable cup on Linux/macOS.
• build/header_only/cup.h.

Codex Location

Install this folder as codex-cup-release/SKILL.md under Codex's configured skills directory, such as $CODEX_HOME/skills or ~/.codex/skills.

First Steps

From the consumer project's root directory, detect how Cup is available:

1. If cup.h exists in the current directory, use header-only mode.
2. Otherwise, if cup.exe exists on Windows, use embedded mode.
3. Otherwise, on Linux/macOS, if an executable named cup or starting with cup can be run as ./cup..., use embedded mode.
4. If neither is present, ask for the release artifact. If the user has build/embedded/cup.exe, run it from the project root or place it in the project root as cup.exe. If the user has build/header_only/cup.h, place it in the project root as cup.h and compile it.

If both cup.h and a Cup executable are present, treat the project as header-only mode.

Cup build scripts are C files named build.c. In both embedded and header-only release usage, build scripts include:

#include "cup.h"

Do not include Cup source-tree paths such as cup/cup.h unless the user's project explicitly provides that layout.

Run Cup

Embedded mode on Windows:

.\cup.exe -hh
.\cup.exe

Embedded mode on Linux/macOS:

./cup -hh
./cup

Header-only mode needs a one-time compile of cup.h with MAIN_ENTRY.

Windows with MSVC:

cl /Fe:cup.exe /Tc cup.h /std:clatest /DMAIN_ENTRY /nologo
.\cup.exe -hh

Linux:

clang -x c cup.h -DMAIN_ENTRY -o cup -D_GNU_SOURCE -fms-extensions
./cup -hh

macOS:

clang -x c cup.h -DMAIN_ENTRY -o cup -fms-extensions
./cup -hh

Always run cup -hh for the actual artifact in the current project before relying on options. Key options:

• -out_dir <dir>: output directory, default build.
• -t <toolchain>: llvm, msvc, gcc, or zig.
• -linker <linker>: LLVM linker, usually default or lld.
• -O0, -O3, -Os: debug, fast release, size release.
• -clean: remove generated artifacts known to Cup.
• -dry: print/prepare work without executing commands.
• -test [targets...]: build and run test executables.
• -root <dir>: resolve relative paths from another root.

Cup records the last selected toolchain, optimization level, and LLVM linker in {out_dir}/.last_status. If later commands omit -t, -O..., or -linker, Cup reuses those saved values. Pass these flags explicitly when changing configuration or when reproducibility matters. Changing -out_dir changes where this status is read and written.

The CUP_CLANG environment variable overrides the LLVM toolchain's C/C++ compiler path:

export CUP_CLANG=/usr/lib/llvm-21/bin/clang-21
./cup -t llvm           # uses clang-21 / clang++-21

Minimal build.c

Use this shape when creating or repairing a simple project:

#include "cup.h"

ENTRY(build_app)
{
    Node* src = SRC("src/main.c");
    Node* obj = OBJ(src);
    CC(src, obj);

    Node* exe = EXE("{out_dir}/app");
    Node* link = LINK(exe);
    link_cmd_add_input(link, obj);
}

int main(int argc, char** argv)
{
    set_test_enabled(true);
    return execute();
}

Run it:

.\cup.exe
.\cup.exe build/app.exe

On Linux/macOS the executable target is usually build/app, because EXE(...) appends an empty executable extension there.

Macro Rules

These macros are the high-signal surface area for everyday Cup build scripts:

• SRC(fmt, ...): declare a C/C++ source node.
• OBJ(src): get the default object node for a source.
• CC(input, output): create a compile command, usually CC(src, OBJ(src)).
• EXE(fmt, ...): declare an executable output and append {exe_ext}.
• DLL(fmt, ...): declare a shared library output and append {dll_ext}.
• LIB(fmt, ...): declare a static library output and append {lib_ext}.
• FILE(fmt, ...): declare a normal file node, useful for headers, generated files, assets, stamps, and copy inputs/outputs.
• LINK(output): create a linker command for an EXE(...) or DLL(...) output.
• AR(output): create an archive/static-library command for a LIB(...) output.
• CMD(command_line): create a custom external process command.
• CMD_FROM_EXE(exe, name): create a command that runs an executable node built by Cup.
• CALLBACK_CMD(fn, ctx): create an in-process C callback command.
• COPY(src, dst): create a platform-specific copy command.

Mental model:

• SRC, OBJ, EXE, DLL, LIB, and FILE create or fetch nodes. They do not build anything by themselves.
• CC, LINK, AR, CMD, CMD_FROM_EXE, CALLBACK_CMD, and COPY create commands that build or update nodes.
• EXE("{out_dir}/app") only names the executable node; LINK(EXE("{out_dir}/app")) creates the link command.
• LIB("{out_dir}/core") only names the static library node; AR(LIB("{out_dir}/core")) creates the archive command.

Important rule: do not use LINK() to generate a static library. Static libraries are LIB(...) plus AR(...). LINK() is for linked binaries: executables and DLL/shared libraries.

Correct:

Node* lib = LIB("{out_dir}/core");
Node* ar = AR(lib);
ar_cmd_add_input(ar, obj);

Never write LINK(lib) or LINK(LIB(...)) for a static library target.

Common Recipes

Compile one source to one object:

Node* src = SRC("src/foo.c");
Node* obj = OBJ(src);
CC(src, obj);

Build an executable:

Node* src = SRC("src/main.c");
Node* obj = OBJ(src);
CC(src, obj);

Node* link = LINK(EXE("{out_dir}/app"));
link_cmd_add_input(link, obj);

Build a static library:

Node* lib = LIB("{out_dir}/core");
Node* ar = AR(lib);

Node* src_a = SRC("src/a.c");
Node* obj_a = OBJ(src_a);
CC(src_a, obj_a);
ar_cmd_add_input(ar, obj_a);

Node* src_b = SRC("src/b.c");
Node* obj_b = OBJ(src_b);
CC(src_b, obj_b);
ar_cmd_add_input(ar, obj_b);

Build an executable that links objects and a static library:

Node* lib = LIB("{out_dir}/core");
Node* ar = AR(lib);

Node* core_src = SRC("src/core.c");
Node* core_obj = OBJ(core_src);
CC(core_src, core_obj);
ar_cmd_add_input(ar, core_obj);

Node* app_src = SRC("src/main.c");
Node* app_obj = OBJ(app_src);
CC(app_src, app_obj);

Node* link = LINK(EXE("{out_dir}/app"));
link_cmd_add_input(link, app_obj);
link_cmd_add_input(link, lib);

Build a DLL/shared library:

Node* src = SRC("src/plugin.c");
Node* obj = OBJ(src);
CC(src, obj);

Node* link = LINK(DLL("{out_dir}/plugin"));
link_cmd_add_input(link, obj);

Copy a file:

Node* input = FILE("assets/config.json");
Node* output = FILE("{out_dir}/config.json");
COPY(input, output);

Run a generated tool:

Node* gen = EXE("{out_dir}/gen_assets");
Node* run = CMD_FROM_EXE(gen, "generate assets");
cmd_add_output(run, FILE("{out_dir}/assets.c"));

Build Targets

Cup accepts zero, one, or many target paths after options:

.\cup.exe
.\cup.exe build/core.lib
.\cup.exe build/core.lib build/app.exe
.\cup.exe -t llvm -O3 build/core.lib

Targets are node paths relative to the current root directory. Use Linux-style / separators even on Windows: build/core.lib, not build\core.lib.

The target path must match a node created by build.c. Examples:

• EXE("{out_dir}/app") becomes build/app.exe on Windows and build/app on Linux/macOS.
• DLL("{out_dir}/plugin") becomes build/plugin.dll on Windows and build/plugin.so on Linux.
• LIB("{out_dir}/core") becomes build/core.lib on Windows and build/core.a on Linux/macOS.
• FILE("{out_dir}/generated/config.h") becomes build/generated/config.h.

Use -dry when unsure:

.\cup.exe -dry build/core.lib

Tests

Test sources include cup/test.h and declare test cases with TEST(fn_name, group_name) or TEST(fn_name).

#include "cup/test.h"

TEST(test_parser_accepts_empty_file, parser)
{
    ASSERT(true);
}

Compile test sources in ENTRY blocks:

#include "cup.h"

ENTRY(build_test_parser)
{
    Node* src = SRC("{dir}/test_parser.c");
    Node* obj = OBJ(src);
    CC(src, obj);
    obj_add_link_obj_from_src(obj, SRC("src/parser.c"));
}

int main(int argc, char** argv)
{
    set_test_enabled(true);
    add_build_script("tests/build.c");
    return execute();
}

Run all tests:

.\cup.exe -test

Run one or more test executables by target path:

.\cup.exe -test build/tests/tests/xxx.c.exe
.\cup.exe -test build/tests/tests/aaa.c.exe build/tests/tests/bbb.c.exe

On Linux/macOS, test executable targets use .test:

./cup -test build/tests/tests/xxx.c.test

How Cup derives the test target path:

1. -test enables test scanning.
2. Cup scans compiled sources for TEST(...).
3. For each compiled source with tests, Cup calls add_test_exe_for_obj(obj, entries).
4. The source path comes from the object's compile command.
5. Windows test exe path: {out_dir}/tests/<source-path>.exe.
6. Linux/macOS test exe path: {out_dir}/tests/<source-path>.test.

Example: if tests/build.c compiles SRC("{dir}/xxx.c"), {dir} resolves to tests, so the source path is tests/xxx.c. With default {out_dir} of build, the Windows test target is build/tests/tests/xxx.c.exe. The repeated tests/tests is intentional: the first tests is Cup's test-output directory, and the second is the source path's tests/ directory.

ENTRY And DAG Edits

All build graph changes happen inside ENTRY functions. Add source, object, command, output, and dependency nodes there.

Use {dir} for paths relative to the build.c file that contains the current ENTRY. Use {out_dir} for generated outputs.

Register extra build scripts in main:

int main(int argc, char** argv)
{
    set_test_enabled(true);
    add_build_script("src/build.c");
    add_build_script("tests/build.c");
    return execute();
}

Use priorities when an entry must run before or after normal node creation:

ENTRY(set_project_vars, PRIORITY_BEFORE_DEFAULT)
{
    set_var("generated_dir", "{out_dir}/generated");
}

ENTRY(adjust_nodes, PRIORITY_AFTER_DEFAULT)
{
    /* Inspect or adjust nodes after normal entries created them. */
}

If all nodes need compile/link flag adjustments, prefer an after_prepare callback. Do not create new targets there; dependencies have already been determined.

static void setup_flags(void)
{
    Node** nodes = get_all_nodes();
    for (size_t i = 0; i != array_size(nodes); i++)
    {
        if (nodes[i]->type == node_make_cmd_type(CMD_TYPE_EXECUTABLE, C_CMD_COMPILE))
        {
            c_compile_cmd_add_include_directory(nodes[i], "src");
        }
    }
}

int main(int argc, char** argv)
{
    set_after_prepare_callback(setup_flags);
    return execute();
}