Bulding Zsh-Timetracker:

Learning Z shell scripting with ChatGPT

• Project Repo

While tracking hours for a contracting gig using Google Sheets, I got curious: could I write a script to do this for me? A classic early-career developer's paradigm.

I took this as an opportunity to try two new things:

• Learn shell scripting
• Utilize ChatGPT as a learning/productivity tool

The first iteration of the program was produced from a detailed ChatGPT prompt, while subsequent iterations were developed through trial and error, further prompts, and reading documentation resources.

ChatGPT produces accurate and readable explanations of core function and syntax issues, and frequently produced usable solutions to certain problems. I was exposed to a LOT of core concepts in shell scripting, such as expansion, conditional expressions, command substitution and parameters

However, the initial design solutions proposed by ChatGPT where rarely optimized for the project scope, and frequently took me in multiple directions before I settled on a solution.

CHALLENGES AND SOLUTIONS

• Relative vs absolute filepaths. Timetracker data is stored in .csv file. I wanted users to be able to run timetracker from any file location on their system. After several attempts, I found a solution utilizing the readlink and dirname $0 commands. read about it here.

# readlink $0 and dirname together locate the
# absolute path to the .csv file even when
# users are accessing timetracker via a symlink.

script_path="$(readlink -f "$0")"
script_dir="$(dirname "$script_path")"
CSV_FILE="${script_dir}/timetracker.csv"

• User inputs. ChatGPT's first suggested script architecture was a convluted set of if... else statements. I refactored the initial output so that the main script of Timetracker is a while... loop that calls functions based on user commands.

# abbreviated example:
while true; do
  read command
  case $command in
    "start")
      cmd_start ;;
    "stop")
      cmd_stop ;;
    "-p "* || "-t "* || "-pt "* )
      cmd_set_proj "$command" ;;
  esac
done

• Parameter expansion and user inputs. At, I did not understand the difference between command-line arguments and user inputs. The initial outputs assumed the program would be running from the command-line, and proposed solutions like the getopts utility. This was not suited for my needs. Eventually, I settled on paramete expansion to break apart user input tags -p, -t from their associated arguments.

# declares a top level associative array PROJECT

declare -A PROJECT=(prj "" task "" start_time "" stop_time "")

# local ARGS=(${(s: :)1}) creates a local variable array args
# by separating user input string on whitespace.
# if user input is '-pt NEWPROJECT BRAINSTORM',
# args=('-pt' 'NEWPROJECT' 'BRAINSTORM')
# cmd_set_proj sets project and task names to PROJECT array.

cmd_set_proj () {
  local ARGS=(${(s: :)1})
  case $ARGS[1] in
  -pt)
    PROJECT[prj]=$args[2]
    PROJECT[task]=$args[3] ;;
# ...
  esac
}

TAKEAWAYS

ChatGPT was an effective tool, but my use of it was often inefficient. Given my inexperience with Unix, I was destined to confront a learning curve. However, I founnd myself learning core shell concepts at the same time as problem-solving issues with ChatGPT-generated program designs solutions.

If I pursue this method in the future, I will spend more time on project design before submitting prompts to ChatGPT.