Editor’s note: This article was last reviewed on 12 July 2024 by Eze Sunday and updated to include new information in more recent Clap versions, advanced use cases like using derive proc macros, a comparison to other CLI parsing libraries, and more.
A CLI application is a program that users interact with through text commands in a terminal. An example is cargo run
, which runs a Rust application bootstrapped with Cargo.
In this article, we will see how to manually parse command line arguments in a Rust application, why manual parsing might not be a good choice for larger apps, and how the Clap library helps solve these issues.
As a note, you should be comfortable reading and writing basic Rust, such as variable declarations, if...else
blocks, loops, and structs.
Let’s say we have a projects folder with many Node-based projects, and we want to find out which packages we’ve used — including dependencies — and how many times we’ve used them. After all, that combined 1+GB of node_modules
can’t be all unique dependencies, right? 😰
What if we made a nice little program that counts the number of times we use a package in our projects?
To do this, let’s set up a project with a cargo new package-hunter
in Rust. The src/main.rs
file is the default main function and will have this code:
fn main() { println!("Hello, world!"); }
We’ll replace the default file content with our code. First, let’s define a function to get the arguments from the user and call the function in the main function like so:
fn get_arguments() { let args: Vec<_> = std::env::args().collect(); // get all arguments passed to app println!("{:?}", args); } fn main() { get_arguments(); }
When we run the above code, we’ll get a nice output that shows all the arguments in an array without errors or panic:
# anything after '--' is passed to your app, not to cargo > cargo run -- svelte Finished dev [unoptimized + debuginfo] target(s) in 0.01s Running `target/debug/package-hunter svelte` ["target/debug/package-hunter", "svelte"] //<- arguments
The first argument is the path to the executable while the second argument is the parameter passed to the executable.
Now that we have the basics set up for our CLI, we can now merrily go on to write the function to count the dependencies. It will take a name
, and counts directories that match that name
in the subdirectories. We’ll use vecDeque
, fs
, and PathBuff
from the standard library to process the directory structure:
use std::collections::VecDeque; use std::fs; use std::path::PathBuf; /// Not the dracula fn count(name: &str) -> std::io::Result<usize> { let mut count = 0; // queue to store next dirs to explore let mut queue = VecDeque::new(); // start with current dir queue.push_back(PathBuf::from(".")); loop { if queue.is_empty() { break; } let path = queue.pop_back().unwrap(); for dir in fs::read_dir(path)? { // the for loop var 'dir' is actually a result, so we convert it // to the actual dir struct using ? here let dir = dir?; // consider it only if it is a directory if dir.file_type()?.is_dir() { if dir.file_name() == name { // we have a match, so stop exploring further count += 1; } else { // not a match so check its sub-dirs queue.push_back(dir.path()); } } } } return Ok(count); }
That’s a pretty long code block. Let’s break it down.
In the above code, we import VecDeque
for efficient queue operations, fs
for filesystem operations, and PathBuf
for handling filesystem paths. Here’s the part of the code block I’m talking about:
use std::collections::VecDeque; use std::fs; use std::path::PathBuf;
For each directory entry in the current path, we:
?
name
, increment the count
Here’s the code I’m talking about:
for dir in fs::read_dir(path)? { let dir = dir?; if dir.file_type()?.is_dir() { if dir.file_name() == name { count += 1; } else { queue.push_back(dir.path()); } } }
We’ll update the get_arguments
function to return the first argument after the command:
fn get_arguments() -> String { let args: Vec<_> = std::env::args().collect(); args[1].clone() }
And in main
, we call count
with that argument:
fn main() { let args = get_arguments(); match count(&args) { Ok(c) => println!("{} uses found", c), Err(e) => eprintln!("error in processing : {}", e), } }
When we run this inside one of the project folders, it unexpectedly works perfectly and returns the count as 1
because a single project will contain a dependency only once:
Now, as we go a directory up, and try to run it, we notice a problem: it takes a little more time because there are more directories to go through.
Ideally, we want to run it from the root of our project directory, so we can find all the projects that have that dependency, but this will take even more time.
So, we decide to compromise and only explore directories until a certain depth. If the depth of a directory is more than the depth given, it will be ignored. Update these sections of the sections of the code below to support the depth:
/// Not the dracula fn count(name: &str, max_depth: usize) -> std::io::Result<usize> { ... queue.push_back((PathBuf::from("."), 0)); ... let (path, crr_depth) = queue.pop_back().unwrap(); if crr_depth > max_depth { continue; } ... // not a match so check its sub-dirs queue.push_back((dir.path(), crr_depth + 1)); ... }
Now the application takes in two parameters: first the package name, then the maximum depth to explore.
However, we want the depth to be an optional argument, so if not given, it will explore all the subdirectories, else it will stop at the given depth.
For this, we can update the get_arguments
function to make the second argument optional:
fn get_arguments() { let args: Vec<_> = std::env::args().collect(); let mdepth = if args.len() > 2 { args[2].parse().unwrap() } else { usize::MAX }; println!("{:?}", count(&args[1], mdepth)); }
With this, we can run it in both ways, and it works:
> cargo run -- svelte > cargo run -- svelte 5
Unfortunately, this is not very flexible. When we give the arguments in reverse order, like cargo run 5 package-name
, the application crashes as it tries to parse package-name
as a number.
Now, we might want the arguments to have their flags, something like -f
and -d
so we can give them in any order. (Also bonus Unix points for flags!)
We again update the get_arguments
function, and this time add a proper struct for the arguments, so returning the parsed arguments is easier:
#[derive(Default)] struct Arguments { package_name: String, max_depth: usize, } fn get_arguments() -> Arguments { let args: Vec<_> = std::env::args().collect(); // atleast 3 args should be there : the command name, the -f flag, // and the actual file name if args.len() < 3 { eprintln!("filename is a required argument"); std::process::exit(1); } let mut ret = Arguments::default(); ret.max_depth = usize::MAX; if args[1] == "-f" { // it is file ret.package_name = args[2].clone(); } else { // it is max depth ret.max_depth = args[2].parse().unwrap(); } // now that one argument is parsed, time for seconds if args.len() > 4 { if args[3] == "-f" { ret.package_name = args[4].clone(); } else { ret.max_depth = args[4].parse().unwrap(); } } return ret; } fn count(name: &str, max_depth: usize) -> std::io::Result<usize> { ... } fn main() { let args = get_arguments(); match count(&args.package_name, args.max_depth) { Ok(c) => println!("{} uses found", c), Err(e) => eprintln!("error in processing : {}", e), } }
Now, we can run it with fancy -
flags like cargo run -- -f svelte
or cargo run -- -d 5 -f svelte
.
However, this has some pretty serious bugs: we can give the same argument twice, and thus skip the file argument entirely cargo run -- -d 5 -d 7
, or we can give invalid flags and this runs without any error message 😭
We can fix this by checking that the file_name
is not empty on line 27
above, and possibly printing what is expected when incorrect values are given. But, this also crashes when we pass a non-number to -d
, as we directly call unwrap
on parse
.
Also, this application can be tricky for new users because it does not provide any help information. Users might not know what arguments will pass and in which order, and the application does not have an -h
flag, like conventional Unix programs, to display that information.
Even though these are just little inconveniences for this specific app, as the number of options will grow as their complexity increases, it becomes harder and harder to maintain all of this manually.
This is where Clap comes in.
Clap is a library that provides functionality to generate parsing logic for arguments, and provides a neat and tidy CLI for applications, including an explanation of arguments, and an -h
help command.
Using Clap is pretty easy, and requires only minor changes to our current setup.
Clap has three common versions used in many Rust projects: v2, v3 and v4. v2 primarily provides a builder-based implementation for building a command line argument parser.
Clap v4 is the most recent major release (at the time of this writing), which builds on the existing features of v3 derive
proc-macros along with the builder implementation. So, we can annotate our struct, and the macro will derive the necessary functions for us.
Both of these have their benefits, and for a more detailed differences and features list, we can check out their documentation and help pages, which provide examples and suggest which situations derive and builder are suitable for.
In this post, we will see how to use Clap v4 with the proc-macro.
To incorporate Clap into our project, add the following in the Cargo.toml
:
[dependencies] clap = { version = "4.0", features = ["derive"] }
This adds Clap as a dependency with its derive
features.
Now, let’s remove the get_arguments
function and its call from main
:
use clap::{Parser, Subcommand}; use std::collections::VecDeque; use std::fs; use std::path::PathBuf; #[derive(Default)] struct Arguments { package_name: String, max_depth: usize, } /// Not the dracula fn count(name: &str, max_depth: usize, logger: &logger::DummyLogger) -> std::io::Result<usize> { let mut count = 0; logger.debug("Initializing queue"); // queue to store next dirs to explore let mut queue = VecDeque::new(); logger.debug("Adding current dir to queue"); // start with current dir queue.push_back((PathBuf::from("."), 0)); logger.extra("starting"); loop { if queue.is_empty() { logger.extra("queue empty"); break; } let (path, crr_depth) = queue.pop_back().unwrap(); logger.debug(format!("path :{:?}, depth :{}", path, crr_depth)); if crr_depth > max_depth { continue; } logger.extra(format!("exploring {:?}", path)); for dir in fs::read_dir(path)? { let dir = dir?; // we are concerned only if it is a directory if dir.file_type()?.is_dir() { if dir.file_name() == name { logger.log(format!("match found at {:?}", dir.path())); // we have a match, so stop exploring further count += 1; } else { logger.debug(format!("adding {:?} to queue", dir.path())); // not a match so check its sub-dirs queue.push_back((dir.path(), crr_depth + 1)); } } } } logger.extra("search completed"); return Ok(count); } fn main() {}
Next, in derive
for the Arguments
structure, add Parser
and Debug
:
use clap::Parser; #[derive(Parser,Default,Debug)] struct Arguments {...}
Finally, in main
, call the parse method:
let args = Arguments::parse(); println!("{:?}", args);
If we run the application with cargo run
, without any arguments, we get an error message:
error: The following required arguments were not provided: <PACKAGE_NAME> <MAX_DEPTH> USAGE: package-hunter <PACKAGE_NAME> <MAX_DEPTH> For more information try --help
This is already better error reporting than our manual version!
As a bonus, it automatically provides an -h
flag for help, which can print the arguments and their order:
package-hunter USAGE: package-hunter <PACKAGE_NAME> <MAX_DEPTH> ARGS: <PACKAGE_NAME> <MAX_DEPTH> OPTIONS: -h, --help Print help information
And now, if we provide something other than a number for MAX_DEPTH
, we get an error saying the string provided is not a number:
> cargo run -- 5 test error: Invalid value "test" for '<MAX_DEPTH>': invalid digit found in string For more information try --help
If we provide them in the correct order, we get the output of println
:
> cargo run -- test 5 Arguments { package_name: "test", max_depth: 5 }
All of this with just two new lines and no need to write any parsing code or do any error handling! 🎉
Currently, our help message is a bit bland because it only shows the argument’s name and order. It would be more helpful to users if they could see what a particular argument is meant for, maybe even the application version in case they want to report any error.
Clap also provides options for this:
#[derive(Parser, Debug)] #[command(author = "Author Name", version, about="A Very simple Package Hunter")] struct Arguments{...}
Now, the -h
output shows all the details and also provides a -V
flag to print out the version number:
package-hunter 0.1.0 Author Name A Very simple Package Hunter USAGE: package-hunter <PACKAGE_NAME> <MAX_DEPTH> ARGS: <PACKAGE_NAME> <MAX_DEPTH> OPTIONS: -h, --help Print help information -V, --version Print version information
It can be a bit tedious to write multiple lines about information in the macro itself. So instead, we can add a doc comment using ///
for the struct, and the macro will use it as the about information. In case both are present, the one in macro takes precedence over the doc comment:
#[command(author = "Author Name", version, about)] /// A Very simple Package Hunter struct Arguments {...}
This provides the same help as before.
To add information about the arguments, we can add similar comments to the arguments themselves:
package-hunter 0.1.0 Author Name A Very simple Package Hunter USAGE: package-hunter <PACKAGE_NAME> <MAX_DEPTH> ARGS: <PACKAGE_NAME> Name of the package to search <MAX_DEPTH> maximum depth to which sub-directories should be explored OPTIONS: -h, --help Print help information -V, --version Print version information
This is much more helpful!
Now, let us bring back the other features we had, such as argument flags (-f
and -d
) and setting the depth argument optional.
Clap makes flag arguments ridiculously simple: we simply add another Clap macro annotation to the struct member with #[
arg(short, long)]
.
Here, short
refers to the shorthand version of the flag, such as -f
, and long
refers to the complete version, such as --file
. We can choose either or both. With this addition, we now have the following:
package-hunter 0.1.0 Author Name A Very simple Package Hunter USAGE: package-hunter --package-name <PACKAGE_NAME> --max-depth <MAX_DEPTH> OPTIONS: -h, --help Print help information -m, --max-depth <MAX_DEPTH> maximum depth to which sub-directories should be explored -p, --package-name <PACKAGE_NAME> Name of the package to search -V, --version Print version information
With both the arguments having flags, there are now no positional arguments; this means we cannot run cargo run -- test 5
because Clap will look for the flags and give an error that the arguments are not provided.
Instead, we can run cargo run -- -p test -m 5
or cargo run -- -m 5 -p test
and it will parse both correctly, giving us this output:
Arguments { package_name: "test", max_depth: 5 }
Because we always need the package name, we can make it a positional argument so we don’t need to type the -p
flag each time.
To do this, remove the #[
arg(short,long)]
from it; now the first argument without any flags will be considered as package name
:
> cargo run -- test -m 5 Arguments { package_name: "test", max_depth: 5 } > cargo run -- -m 5 test Arguments { package_name: "test", max_depth: 5 }
One thing to note in shorthand arguments is that if two arguments begin with the same letter — package-name
and path
— and both have a short flag enabled. The application will crash at runtime for debug builds and give some confusing error messages for release builds.
So, make sure that either:
short
flagThe next step is to make the max_depth
optional.
To mark any argument as optional, simply make that argument’s type Option<T>
where T
is the original type argument. So in our case, we have the following:
#[arg(short, long)] /// maximum depth to which sub-directories should be explored max_depth: Option<usize>,
This should do the trick. The change is also reflected in the help, where it does not list the max depth as a required argument:
package-hunter 0.1.0 Author Name A Very simple Package Hunter USAGE: package-hunter [OPTIONS] <PACKAGE_NAME> ARGS: <PACKAGE_NAME> Name of the package to search OPTIONS: -h, --help Print help information -m, --max-depth <MAX_DEPTH> maximum depth to which sub-directories should be explored -V, --version Print version information
And, we can run it without giving the -m
flag:
> cargo run -- test Arguments { package_name: "test", max_depth: None }
But, this is still a little cumbersome; now we must run match
on max_depth
, and if it is None
, we set it to usize::MAX
as before.
Clap, however, has something for us here as well! Instead of making it Option<T>
, we can set the default value of an argument if not given.
So after modifying it like this:
#[arg(default_value_t=usize::MAX,short, long)] /// maximum depth to which sub-directories should be explored max_depth: usize,
We can run the application with or without providing the value of max_depth
(the max value for usize
depends on your system configuration):
> cargo run -- test Arguments { package_name: "test", max_depth: 18446744073709551615 } > cargo run -- test -m 5 Arguments { package_name: "test", max_depth: 5 }
Now, let’s hook it up to the count function in main
like before:
fn main() { let args = Arguments::parse(); match count(&args.package_name, args.max_depth) { Ok(c) => println!("{} uses found", c), Err(e) => eprintln!("error in processing : {}", e), } }
And with this, we have our original functionality back, but with much less code and some extra added features!
The package-hunter
is performing as expected, but alas, there is a subtle bug that has been there since the manual parsing stage and carried to the Clap-based version. Can you guess what it is?
Even though it is not a very dangerous bug for our tiny app, it can be a vulnerability for other applications. In our case, it will give a false result when it should give an error.
Try running the following :
> cargo run -- "" 0 uses found
Here, the package_name
is passed in as an empty string when an empty package name should not be allowed. This happens due to the way the shell we run the command from passes the arguments to our app.
Usually, the shell uses spaces to split the argument list passed to the program, so abc def hij
will be given as three separate arguments: abc
, def
, and hij
.
If we want to include the space in an argument, we must put quotes around it, like "abc efg hij"
. That way the shell knows this is a single argument and it passes it as such.
On the other hand, this also allows us to pass empty strings or strings with only spaces to the app. Again, Clap to the rescue! It provides a way to deny empty values for an argument:
#[arg(value_parser = validate_package_name)] /// Name of the package to search package_name: String,
With this, if we try to give an empty string as the argument, we get an error:
> cargo run -- "" error: The argument '<PACKAGE_NAME>' requires a value but none was supplied
But, this still provides spaces as a package name, meaning ""
is a valid argument. To fix this, we must provide a custom validator to check if the name has any leading or trailing spaces and reject it if it does.
We define our validation function as the following:
fn validate_package_name(name: &str) -> Result<(), String> { if name.trim().len() != name.len() { Err(String::from( "package name cannot have leading and trailing space", )) } else { Ok(()) } }
Then, set it up for package_name
as follows:
#[arg(value_parser = validate_package_name)] /// Name of the package to search package_name: String,
Now, if we try to pass an empty string or string with spaces, it will give an error, as it should:
> cargo run -- "" error: The argument '<PACKAGE_NAME>' requires a value but none was supplied > cargo run -- " " error: Invalid value " " for '<PACKAGE_NAME>': package name cannot have leading and trailing space
This way, we can validate the arguments with a custom logic without writing all the code for parsing it.
The application is working fine now, but we have no way to see what happened in the cases when it didn’t. For that, we should keep logs of what our application is doing to see what happened when it crashed.
Just like other command line applications, we should have a way for users to set the level of the logs easily. By default, it should only log major details and errors so the logs aren’t cluttered, but in cases when our application crashes, there should be a mode to log everything possible.
Like other applications, let’s make our app take the verbosity level using a -v
flag; no flag is the minimum logging, -v
is intermediate logging, and -vv
is maximum logging.
To do this, Clap provides a way so that the value of an argument is set to the number of times it occurs, which is exactly what we need here! We can add another parameter, and set it as the following:
#[arg(short, long, parse(from_occurrences))] verbosity: usize,
Now, if we run it without giving it a -v
flag, it will have a value of zero, and otherwise count how many time -v
flag occurs:
> cargo run -- test Arguments { package_name: "test", max_depth: 18446744073709551615, verbosity: 0 } > cargo run -- test -v Arguments { package_name: "test", max_depth: 18446744073709551615, verbosity: 1 } > cargo run -- test -vv Arguments { package_name: "test", max_depth: 18446744073709551615, verbosity: 2 } > cargo run -- -vv test -v Arguments { package_name: "test", max_depth: 18446744073709551615, verbosity: 3 }
Using this value, we can easily initialize the logger and make it log the appropriate amount of details.
I have not added the dummy logger code here, as this post focuses on the argument parsing, but you can find it in the repository at the end.
Now that our application is working well, we want to add another functionality: listing the projects we have. That way, when we want a nice list of projects, we can quickly get one.
Clap has a powerful subcommand feature that can provide apps with multiple subcommands. To use it, define another struct with its own arguments, the subcommand. The main argument struct contains the arguments common to all the subcommands, and then the subcommands.
We will structure our CLI as the following:
max_depth
parameters will be in the main structureprojects
command takes an optional start path to start the searchprojects
command takes an optional exclude paths list, which skips the given directoriesThus, we add the count and project enum as below:
use clap::{Parser, Subcommand}; ... #[derive(Subcommand, Debug)] enum SubCommand { /// Count how many times the package is used Count { #[arg(value_parser = validate_package_name)] /// Name of the package to search package_name: String, }, /// list all the projects Projects { #[arg(short, long, default_value_t = String::from("."), value_parser = validate_package_name)] /// directory to start exploring from start_path: String, #[arg(short, long, value_delimiter = ':')] /// paths to exclude when searching exclude: Vec<String>, }, }
Here, we move the package_name
to the Count
variant and add the start_path
and exclude
options in the Projects
variant.
Now, if we check help, it lists both of these subcommands and each of the subcommand has its own help.
Then we can update the main function to accommodate them:
fn main() { let args = Arguments::parse(); let logger = logger::DummyLogger::new(args.verbosity as usize); match args.cmd { SubCommand::Count { package_name } => match count(&package_name, args.max_depth, &logger) { Ok(c) => println!("{} uses found", c), Err(e) => eprintln!("error in processing : {}", e), }, SubCommand::Projects { start_path, exclude, } => match projects(&start_path, args.max_depth, &exclude, &logger) { Ok(_) => {} Err(e) => eprintln!("error in processing : {}", e), }, } }
We can also use the count
command like before to count the number of uses:
> cargo run -- -m 5 count test
As max_depth
is defined in the main Arguments
struct, it must be given before the subcommand.
We can then give multiple values to the project’s command’s excluded directories, as needed:
> cargo run -- projects -e ./dir1 ./dir2 ["./dir1", "./dir2"] # value of exclude vector
We can also set a custom separator, in case we don’t want the values to be separated by space, but by a custom character:
#[arg(short, long, multiple_values = true, value_delimiter = ':')] /// paths to exclude when searching exclude: Vec<String>,
Now we can use :
to separate values:
> cargo run -- projects -e ./dir1:./dir2 ["./dir1", "./dir2"]
This completes the CLI for the application. The project listing function is not shown here, but you can try writing that on your own or check its code in the GitHub repository.
There are a few other libraries that allow you to build command-line parsing applications. They include Argh, Pico-args, and Gumdrop. Here is a summary of their comparison based on their popularity, documentation, maintenance, etc.:
Argh | Clap | Pico-args | Gumdrop | |
---|---|---|---|---|
Community | Small, growing (1.6k stars on GitHub) | Large, active (13.7k stars on GitHub) | Small, dedicated (555 stars) | Very small (223 github stars) |
Docs | Basic examples on GitHub | Rust doc with an example repository | Rust style doc | Rust style doc |
Ease of use | Easy to use | Easy to use | Easy to use | Moderate setup, straightforward |
Customizability | Limited and straightforward | Highly customizable | Minimal customization options | limited and straight forward |
Subcommand support | Yes | Yes | Yes | Yes |
Dependencies | 2 dependencies and 2 dev dependencies | 2 dependencies and 7 dev dependencies | 0 dependencies | 1 dependency and 1 dev dependency |
Maintenance | Last update was 3 months ago | Last update was a few days ago | Last update was 9 months ago | Most recent update was 2 years ago |
Now that you know about Clap, you can make clean and elegant CLIs for your projects. It has many other features, and if your project needs a specific functionality for the command line, there is a good chance that Clap already has it.
You can check out the Clap docs and GitHub page to see more information on the options that the Clap library provides.
You can also get the code for this project here. Thank you for reading!
Debugging Rust applications can be difficult, especially when users experience issues that are hard to reproduce. If you’re interested in monitoring and tracking the performance of your Rust apps, automatically surfacing errors, and tracking slow network requests and load time, try LogRocket.
LogRocket is like a DVR for web and mobile apps, recording literally everything that happens on your Rust application. Instead of guessing why problems happen, you can aggregate and report on what state your application was in when an issue occurred. LogRocket also monitors your app’s performance, reporting metrics like client CPU load, client memory usage, and more.
Modernize how you debug your Rust apps — start monitoring for free.
Would you be interested in joining LogRocket's developer community?
Join LogRocket’s Content Advisory Board. You’ll help inform the type of content we create and get access to exclusive meetups, social accreditation, and swag.
Sign up nowBuild scalable admin dashboards with Filament and Laravel using Form Builder, Notifications, and Actions for clean, interactive panels.
Break down the parts of a URL and explore APIs for working with them in JavaScript, parsing them, building query strings, checking their validity, etc.
In this guide, explore lazy loading and error loading as two techniques for fetching data in React apps.
Deno is a popular JavaScript runtime, and it recently launched version 2.0 with several new features, bug fixes, and improvements […]