FuncGraft

Ultra lightweight CLI code editor. I tried codex but it felt too bloated, led to code sprawl, and did not feel targeted enough for me. So I built my own equivalent that operates on individual files.

🚫 This code was not AI generated. I do believe in leveraging AI tools for targeted code manipulations, but I do not believe in letting AI do all the work.

Table of Contents

• Prerequisites
• Setup
• Usage

Prerequisites

Ensure you possess a valid OpenAI API key. Set it as an environment variable:

export OPENAI_API_KEY="<your-api-key>"

This program requires CMake, {fmt} and libcurl. These can be installed as follows:

Ubuntu/Debian

apt install cmake libfmt-dev libcurl4-openssl-dev

macOS

brew install cmake fmt
# libcurl usually comes bundled with macOS

Other systems

This program should work on other Unix-like systems (i.e. other Linux distributions) however I do not extensively test these.

Setup

Compile the binary by executing the default make target:

make

The binary will be installed into the directory specified by CMake's install() function. To clean up generated artifacts:

make clean

Usage

Example - read from stdin

Assume foo.cpp uses CamelCase formatting and needs to be converted to snake_case. Run:

edit foo.cpp

Which will begin a basic interactive session. Outline the required changes when prompted:

Input: Convert all code from CamelCase to snake_case.

This will print the updated code to stdout and prompt whether to overwrite foo.cpp. To automatically overwrite foo.cpp, simply run:

edit foo.cpp -o foo.cpp

To save the updated code to a new file, such as bar.cpp, execute:

edit foo.cpp -o bar.cpp

Example - working with complex prompts

Complex multiline prompts can be written into a file and read into the program. For example:

echo "Convert all code from CamelCase to snake_case." > edit.txt && \
edit /tmp/test.py -o /tmp/test.py -f edit.txt

Note

The instructions in edit.txt do not require prompt engineering. When processed, these instructions are combined with additional context and output format specifications to create a complete prompt.

Example - targeted code manipulation

Suppose we have the file example.c:

#include <stdio.h>

void printIntAddr()
{
    const int i = 42;
    printf("Value: %d\n", i);
    printf("Address: %p\n", (void *)&i);
}

void print_float_addr()
{
    const float f = 3.14f;
    printf("Value: %.2f\n", f);
    printf("Address: %p\n", (void *)&f);
}

int main()
{
    print_int_addr();
    print_float_addr();
    return 0;
}

This code will not compile. The case between the two functions that print an integer address are mixed. To both resolve this and minimize token usage, the offending code can be isolated with @@@@@ delimiters:

#include <stdio.h>

@@@@@
void printIntAddr()
{
    const int i = 42;
    printf("Value: %d\n", i);
    printf("Address: %p\n", (void *)&i);
}
@@@@@

void print_float_addr()
{
...

We can then run:

edit example.c

And request a CamelCase to snake_case conversion, thus resolving the missing function definition.

Debugging

Run the program with the -v flag to enable verbosity:

edit foo.cpp -v

This will print out the prompt being sent to OpenAI.

Specify a custom model

Run the program with the -m or --model option:

edit foo.cpp -m <your-model>