generate.md

Generate

The core of generate is ./scripts/generate.sh.

generate.sh can directly generate completion scripts for most commands.

However, the help of some commands is not standardized, or some options or positional parameters in the command require dynamic data, you may need to add scripts to assist generate.sh to complete the job.

You may check out the src for examples.

processing flow

generate.sh internally generates the completion script according to the following processing flow.

                        parse-tab.awk                         parse-script.awk 
help-text(_patch_help) ---------------> table(_patch_table) ------------------> script
              

Let's take aichat as an example.

1. help-text: aichat --help

Use ChatGPT, LocalAI and other LLMs in the terminal.

Usage: aichat [OPTIONS] [TEXT]...

Arguments:
  [TEXT]...  Input text

Options:
  -m, --model <MODEL>        Choose a LLM model
  -r, --role <ROLE>          Choose a role
  -s, --session [<SESSION>]  Create or reuse a session
  -H, --no-highlight         Disable syntax highlighting
  -S, --no-stream            No stream output
      --dry-run              Run in dry run mode
      --info                 Print related information
      --list-models          List all available models
      --list-roles           List all available roles
      --list-sessions        List all available sessions
  -h, --help                 Print help
  -V, --version              Print version

2. table: aichat --help | gawk -f scripts/parse-table.awk

option # -m, --model <MODEL> # Choose a LLM model
option # -r, --role <ROLE> # Choose a role
option # -s, --session [<SESSION>] # Create or reuse a session
option # -H, --no-highlight # Disable syntax highlighting
option # -S, --no-stream # No stream output
option # --dry-run # Run in dry run mode
option # --info # Print related information
option # --list-models # List all available models
option # --list-roles # List all available roles
option # --list-sessions # List all available sessions
option # -h, --help # Print help
option # -V, --version # Print version
argument # [TEXT]... # Input text # 

3. script: aichat --help | gawk -f scripts/parse-table.awk | gawk -f scripts/parse-script.awk

# @option -m --model         Choose a LLM model
# @option -r --role          Choose a role
# @option -s --session       Create or reuse a session
# @flag -H --no-highlight    Disable syntax highlighting
# @flag -S --no-stream       No stream output
# @flag --dry-run            Run in dry run mode
# @flag --info               Print related information
# @flag --list-models        List all available models
# @flag --list-roles         List all available roles
# @flag --list-sessions      List all available sessions
# @flag -h --help            Print help
# @flag -V --version         Print version
# @arg text*                 Input text

_patch* function

When generate.sh <cmd> is executed, it will look for a <cmd>.sh file in the src/ directory and run the _patch_* functions in it.

_patch_help

_patch_help is a hook to provide/patch help text. If omitted, generate.sh runs <cmd> [subcmd]... --help to generate help text.

Here are some use cases for _patch_help:

• curl: execute curl --help all to get help text.
• npm: manually provide help text as the help text generated by npm is too non-standard.
• ssh: use man-page other than help text.

_patch_table

_patch_table is a hook to patch the generated table，usually used to bind option/positional parameter to _choice* functions.

Let's also take aichat as an example.

_patch_table() {
    _patch_table_edit_options \
        '--model;[`_choice_model`]' \
        '--role;[`_choice_role`]' \
        '--session;[`_choice_session`]' \

}

With _patch_table, _choice_* is appended to fourth column.

- option # -r, --role <ROLE> # Choose a role
+ option # -r, --role <ROLE> # Choose a role # [`_choice_role`]

This affects the final generated argc script.

- # @option -r --role          Choose a role
+ # @option -r --role[`_choice_role`]      Choose a role

_choice_* function

_choice_fn will generate dynamic completion candidates.

Let's also take aichat as an example.

Suppose you are typing aichat --role <tab>, when <tab> is pressed, _choice_role bound to --role will run and provide completion data.

In _choice_* function, you can use argc_* variables to easily access option/positional value. see argc-variables for more details.

We also privode some utility functions to make it easier to write _choice_* function.

_choice_fn() {
    echo     abc                              # value only,
    echo -e "def\0"                           # value only, no space
    echo -e "ijk\tDesc"                       # value with description
    echo -e "nop\0\tDesc"                     # value with description, no space
}

run Argcfile.sh

Argcfile.sh provide a number of commands that can help with the development of the completion script.

• print: Print help/table/script, used for debugging _patch_help and _patch_table
• generate: Generate the completion script.
• choice-fn: Run a choice fn directly

argc print aichat -k help        # Run: _patch_help
argc print aichat -k table       # Run: _patch_help | awk -f ./scripts/parse-table.awk | _patch_table
argc print aichat -k script      # Run: _patch_help | awk -f ./scripts/parse-table.awk | _patch_table | awk -f ./scripts/parse-script.awk 

argc generate aichat             # Run: ./scripts/generate.sh -o completions/aichat.sh aichat
argc format aichat               # Run: ./scripts/format.sh src/aichat.sh

argc choice-fn src/aichat.sh _choice_model            # Run _choice_model in src/aichat.sh
argc choice-fn completions/aichat.sh _choice_model    # Run _choice_model in completions/aichat.sh