martin-t / cvars Goto Github PK

• cvars_structs! {} to allow defining whole structs that contain each other
• Useful for configuring multiple weapons/vehicles/items/etc that have the same fields - rec-wars does it manually atm.

Compare normal access in gamecode and stringly-typed access when the user is accessing the config. Small and large amount of cvars.

I've got some structs with fields that must not be changed (would cause crashing and/or freezing), but would like to keep the remaining fields configurable...

Would it be possible to add something like a #[cvars(skip)] attribute, similar to #[serde(skip)]? Maybe check for the latter too?

• Tab - autocompletion if unambiguous, print possibilities if ambiguous
• Write text, up arrow -> search history only for items starting with the text
  • Pg up/down still walks history normally (like zsh)
• cvarlist
• find/search/apropos with multiple words - search for those words appearing in any order
• If starting with space, don't save to history
• History to file
• CTRL+R - fuzzy search in history - possible to use fzf or alternative (atuin/skim)?

Collecting isn't necessary when using quote: e.g.:

tasteful API design @ https://matklad.github.io/2022/03/26/self-modifying-code.html

Check for:

• dead links
• FIXMEs
• typos?

When RecWars console works sanely or RustCycles supports WASM, add a heading "Try it out" right before usage that explains how to open the console and which cvars to try changing. Maybe reduce number of bots?

Should warn, not error if not

Look at what strum does in get_documentation.

Switch to a hashmap or trie for storing the strings, the generated match is huge:

cargo llvm-lines:

  Lines          Copies       Function name
  -----          ------       -------------
  119771 (100%)  4182 (100%)  (TOTAL)
   30067 (25.1%)    1 (0.0%)  rec_wars::cvars::Cvars::set_str
    5892 (4.9%)     1 (0.0%)  rec_wars::cvars::Cvars::get_string
    3083 (2.6%)     1 (0.0%)  rec_wars::rendering::render_viewport

and render_viewport is a huge 700 line function.

Maybe llvm lines are not the best metric but it absolutely does add a lot to compile times (test with SetGetDummy)

Would it be possible to have a pair of set_var and get_var-functions that use an integer-index (parallel to str), obtainable by some get_idx function? Maybe also have a macro that can resolve this at compile-time?

This could be implemented by wrapping the memory-offset (using https://crates.io/crates/memoffset ?) in a CvarHandle<T> {offset: u16} struct with an impl PartialEq, where T is the source-type... thus never exposing the actual offset to the library user.

Also offer an iterator of cvar names (and values?)

Right now doc comments cause warning: unused doc comment / rustdoc does not generate documentation for macro invocations

Aliases would allow users to save typing and one cvar could have multiple names for users coming from different engines used to different names.

Either setup a script to run it locally before releasing or add to CI if possible

Currently we hardcode pub in the generated code, which means making the Cvars struct pub(crate) gives an error about leaking a less visible type in a public interface.

• Built-in - help, cvarlist, search/find/apropos, cmdlist ...
• User-defined - just call a hook function and that returns true if handled? Would be nice to be a bit more declarative so they can be listed in help/cmdlist. Avoid making it overcomplicated - define_command function that takes the help message and the hook?

There are almost no allocations, might be worth seeing if it's possible to make it no_std compatible.

Would have to conditionally remove get_string since String allocates.

Seems to do something very similar, compare features and compile times

Unmaintained but might contain useful ideas of code.