# sfw-tools ## Design In the spirit of Software Tools, the aim is to make components re-usable in three ways: 1. Implement core features as functions, so they can be re-used within Rust. These functions should generally return a `Result` type, so the caller can decide how to deal with the error. 2. Executable commands with a simple interface that typically act as thin wrappers around the library functions, or perhaps combine the library functions in interesting ways. 3. As well designed code that can be copied as repurposed when necessary. A fourth avenue may be explored, which is to adopt the [nushell](https://github.com/rjbs/Sweater) approach to transferring tabular data between commands. For a related project that also follows Software Tools in Rust, and may serve as an interesting comparison, see [Sweater](https://github.com/rjbs/Sweater). A more feature-rich project is [uutils coreutils](https://github.com/uutils/coreutils), which as the name suggests, is a Rust implementation analogous to GNU Coreutils. ### Functional facilities Higher-order-functions (HOFs) are frequently used to reduce code complexity, verbosity, and the risk of errors. Primary examples are `map`, `for_each` (like `map` but effectful), and `fold`. As pointed out in Software Tools, pp 21, *"The best programs are designed in terms of loosely coupled functions that each does a simple task."* Some other references that refelect functional programming values: - page 36, a discussion on `break`: the suggestions also coincide largley with recursive functions. - pages 44-45 discuss defensive programming by guarding control variables with safety checks. In functional programming, such control variables often do not appear, so safety checks are unnecessary due to the usage of HOFs being safe by design. Page 45 also points out that non-voluminous code listings are easier to debug (which I agree with, and a functional style typically enables this), though we also want to warn against making code overly terse. Experience is the best guide in this case. ### Currently Implemented Tools - [x] `cp` - [x] `wc` - [x] `detab` - [x] `entab` - [x] `echo` - [x] `compress` - [ ] `expand` ### Dependencies Since the goal is to make the software both as self-contained and as illustrative as possible, we've tried to rely on very few dependencies. The following exceptions exist: - [fp-core](https://docs.rs/fp-core) This is what one would typically find as part the standard library in a functional language, so we have included it here. Though Rust is functional in a sense — it has lambda functions (i.e. Rust closures) and the stand library has many higher-order functions (HOFs) — its standard library doesn't include traits that are commonly found to be helpful abstracts in functional languages. We will use a few of these where it is particularly illustrative or sensible, but will stick with idiomatic Rust where that is obviously simpler. An interesting note is that filters are the subject of chapter 2 and much of the rest of the book, which are just a particular class of HOFs. - [peeking_take_while](https://docs.rs/peeking_take_while/) A small library that provides the `peeking_take_while` function for `Peekable` iterators. This behaves more of how would would expect for a `take_while` function compared to the standard `take_while` implementation, which will "lose" the first element after a `take_while` streak ends. - [tailcall](https://docs.rs/tailcall) This is a macro that enables tailcall elimination for functions that are tail recursive. In other words, instead of writing loops, we can sometimes just write a function that calls itself. Without this macro, such functions would eventually cause the stack to blow up. - [seahorse](https://docs.rs/seahorse) Seahorse is a minimal argument parser. Judging by some results returned by Google, [clap](https://clap.rs) is far more popular, but has additional dependencies; we are striving for being as portable as possible, so the minimality seemed to line up with that goal. Additionally, Clap doesn't appear to allow passing in argument lists directly, which is useful for maintaining separate commands that build on each other. In any case, argument parsing is only used very late in the application logic, and most of the API could be used without worrying about it. #### Currently unused - [byteorder](docs.rs/byteorder) Library for reading/writing numbers in big-endian and little-endian. This is a somewhat low-level library, but as this is an IO-heavy library of tools, it may make sense to rely on it. - [im](https://docs.rs/im) Immutable data structures that implement structural sharing can be even more performant than `std`'s mutable structures for large data types, and while Rust makes mutation far safer than most languages, mutation can still result in confusion at times, so in the cases where clarity is more important than performance (or performance doesn't matter much, e.g. one-ops), it may be preferable to use immutable data structures. ### Build ### Misc Notes #### Using todo!() to Using `todo!()` from `std::todo` is a helpful way to incrementally develop a feature while still getting feedback from the compiler. [**TODO**: show example] A [caveat](https://github.com/rust-lang/rfcs/issues/3045) is that currently you need code in the function after the `todo!()`, even if it doesn't match the type. For instance, we can use a function like: ```rust pub fn some_num() -> i32 { todo!(); (); } ``` Most beneficial is that `rustc` will warn you if you a `todo!()` is left in your code, since it would result in a panic if that execution path were to occur. #### Rust on nix ```plain nix develop ``` #### Optimizing for size * https://github.com/johnthagen/min-sized-rust Currently, to generate small builds the following commands are required. 1. (only once per environment) Make source code for the standard library available: ```plain rustup component add rust-src --toolchain nightly ``` 2. ```plain cargo +nightly build -Z build-std --target x86_64-unknown-linux-gnu --release ``` 3. (optional) `strip` binary - see links in notes ### Project administration #### Git hooks ##### Cargo-Husky We use [cargo-husky](https://github.com/rhysd/cargo-husky) to keep in line; it enforces several checks with a `pre-push` hook. Sometimes it is a bit restrictive, so if we need to push in-progress work to a branch, we can use `git push --no-verify -u origin feature_branch`. Cargo-husky expects certain files to be at the root of the repository, thus the symlinks. ##### pre-commit We include the following, less stringent checks for pre-commit. ```bash #!/bin/sh # Put in your Rust repository's .git/hooks/pre-commit to ensure you never # breaks rustfmt. # # WARNING: rustfmt is a fast moving target so ensure you have the version that # all contributors have. for FILE in `git diff --cached --name-only`; do if [[ -f "$FILE" ]] && [[ $FILE == *.rs ]] \ && ! rustup run nightly rustfmt --unstable-features \ --skip-children $FILE; then echo "Commit rejected due to invalid formatting of \"$FILE\" file." exit 1 fi done cd Rust/sfw-tools && cargo readme > README.md && git add README.md ``` As can be seen this also generates the README from doc comments in `lib.rs`. License: MPL-2.0