Cargo is a good, reliable way to build programs and libraries in Rust with versioned dependencies. Those of us who have worked with the Wild West practices of C++ development find this particularly soothing, and it's one of the core strengths of the Rust ecosystem.
However, it's not intended to make running little test programs straightforward - you have to
create a project with all the dependencies you wish to play with, and then edit src/main.rs and
do cargo run. A useful tip is to create a src/bin directory containing your little programs
and then use cargo run --bin NAME to run them. But there is a better way; if you have such
a project (say called 'cache') then the following invocation will compile and link
a program against those dependencies (rustc is an unusually intelligent compiler)
$ rustc -L /path/to/cache/target/debug/deps mytest.rs
Of course, you need to manually run cargo build on your cache project whenever new dependencies
are added, or when the compiler is updated.
The runner tool helps to automate this pattern. It also supports snippets, which
are little 'scripts' formatted like Rust documentation examples.
$ cat print.rs
println!("Hello, World!");
$ runner print.rs
Hello, World!
This follows basically the same rules as the doc-test snippets you find in Rust
documentation, so runner allows you to copy those snippets into an editor
and directly run them (I bind 'run' for Rust projects to runner ... in
my favourite editor.)
You can use ? in snippets instead of the ubiquitous and awful unwrap, since the boilerplate
encloses code in a function that returns Result<(),Box<Error+Sync+Send>> which is compatible with
any error return.
A special variable args is available containing any arguments passed to the program:
$ cat hello.rs
println!("hello {}",args[1]);
$ runner hello.rs dolly
hello dolly
You can even - on Unix platforms - add a 'shebang' line to invoke runner:
$ cat hello
#!/usr/bin/env runner
println!("Hello, World!");
$ ./hello
Hello, World!
runner adds the necessary boilerplate and creates a proper Rust program in ~/.cargo/.runner/bin,
prefixed with a prelude, which is initially:
#![allow(unused_imports)]
#![allow(unused_variables)]
#![allow(dead_code)]
#![allow(unused_macros)]
use std::{fs,io,env};
use std::fs::File;
use std::io::prelude::*;
use std::path::{PathBuf,Path};
use std::collections::HashMap;
use std::time::Duration;
use std::thread;
macro_rules! debug {
($x:expr) => {
println!("{} = {:?}",stringify!($x),$x);
}
}After first invocation of runner, this is found in ~/.cargo/.runner/prelude;
you can edit it later with runner --edit-prelude.
debug! saves typing: debug!(my_var) is equivalent to println!("my_var = {:?}",my_var).
As an experimental feature, runner will also do some massaging of rustc errors.
They are usually very good, but involve fully qualified type names.
It reduces std:: references to something simpler.
This is a snippet which a Java programmer would find easy to write - declare that type explicitly, and assume that the important verb is "set":
$ cat testm.rs
let mut map: HashMap<String,String> = HashMap::new();
map.set("hello","dolly");
$ runner testm.rs
error[E0599]: no method named `set` found for type `HashMap<String, String>` in the current scope
--> /home/steve/.cargo/.runner/bin/testm.rs:24:9
|
24 | map.set("hello","dolly");
| ^^^
|
= help: did you mean `get`?
Since we are being very informal with Rust here, it's appropriate that we don't wish the type spelled
out in full glory (as you can see by running with -S):
std::collections::HashMap<std::string::String, std::string::String>.
As you can see, runner is very much about playing with small code snippets. By
default it links the snippet dynamically which is significantly faster.
The static option is much more convenient. You can easily create a static cache with some common crates:
$ runner --add "time json regex"
You can add as many crates if you like - number of available dependencies doesn't
slow down the linker. Thereafter, you may refer to these crates in snippets. Note that
by default, runner uses 2018 edition since 0.4.0.
// json.rs
use json;
let parsed = json::parse(r#"
{
"code": 200,
"success": true,
"payload": {
"features": [
"awesome",
"easyAPI",
"lowLearningCurve"
]
}
}
"#)?;
println!("{}",parsed);And then build statically and run (any extra arguments are passed to the program.)
$ runner -s json.rs
{"code":200,"success":true,"payload":{"features":["awesome","easyAPI","lowLearningCurve"]}}A convenient new feature is "argument lines" - if the first line of json.rs was
//: -s
then any runner arguments specified after "//:" will be merged in with the command-line arguments.
It is now possible to simply invoke using runner json.rs. It's better to keep any special build
instructions in the file itself, and it means that an editor run action bound to runner FILE can be
made to work in all cases.
runner provides various utilities for managing the static cache.
You can say runner --edit to edit the static cache Cargo.toml, and runner --build to
rebuild the cache afterwards. runner update will update all the dependencies in the
cache, and runner update package will update a particular package - follow this
with build as before.
The cache is built for both debug and release mode,
so using -sO you can build snippets in release mode. Documentation is also built
for the cache, and runner --doc will open that documentation in the browser. (It's
always nice to have local docs, especially in bandwidth-starved situations.)
If you want docs for a specific crate NAME, then runner --doc NAME will work.
Remember that the Rust documentation generated has a fast offline searchable
index!
The --crates command also has an optional argument; without arguments it lists all
he crates known to runner, with their versions. With a name, it uses an exact match:
$ runner --crates yansi
yansi = "0.3.4"
You may provide a number of crate names here; if --verbose (-v) is specified
then the dependencies of these crates are also listed.
The -c flag only compiles the program or snippet, and copies it to ~/.cargo/bin.
-r only runs the program, which must have previously been compiled, either
explicitly with -c or implicitly with default operation.
Plain Rust source files (which already have fn main) are of course supported, but you
will need the --extern (-x) flag to bring in any external crates from the static cache.
A useful trick - if you want to look at the Cargo.toml of an already downloaded crate
to find out dependencies and features, then this command will open it for you:
favorite-editor $(runner -P some-crate)/Cargo.toml
There are a few Perl-inspired features. The -e flag compiles and evaluates an
expression. You can use it as an unusually strict desktop calculator:
$ runner -e "10 + 20*4.5"
error[E0277]: the trait bound `{integer}: Mul<{float}>` is not satisfied
--> temp/tmp.rs:20:22
|
20 | let res = 10 + 20*4.5;
| ^ no implementation for `{integer} * {float}`
Likewise, you have to say 1.2f64.sin() because 1.2 has ambiguous type.
(Note that the trait std::ops::Mul is presented in simplified form by default)
--expression is very useful if you quickly want to find out how Rust
will evaluate an expression - we do a debug print for maximum flexibility.
$ runner -e 'PathBuf::from("bonzo.dog").extension()'
Some("dog")
(This works because we have a use std::path::PathBuf in the runner prelude.)
Now, this will not work on Windows since quoting
is seriously baroque. So runner re-uses an old trick that some Windows versions of AWK used. We can
only use double-quotes for an argument that may contain spaces, but single-quotes within this will
be converted to double-quotes.
c:> runner -e "PathBuf::from('bonzo.dog').extension()"
Some("dog")
So, in these examples where you need to quote strings in the Rust expression, remember that it works the other way in Windows.
-i (or --iterator) evaluates iterator expressions and does a debug
dump of the results:
$ runner -i '(0..5).map(|i| (10*i,100*i))'
(0, 0)
(10, 100)
(20, 200)
(30, 300)
(40, 400)
Any extra command-line arguments are available for these commands, so:
$ runner -i 'env::args().enumerate()' one 'two 2' 3
(0, "/home/steve/.cargo/.runner/bin/tmp")
(1, "one")
(2, "two 2")
(3, "3")
And finally -n (or --lines) evaluates the expression for each line in
standard input:
$ echo "hello there" | runner -n 'line.to_uppercase()'
"HELLO THERE"
The -x flag (--extern) allows you to insert an extern crate into your
snippet. This is particularly useful for these one-line shortcuts. For
example, my easy-shortcuts crate has a couple of helper functions. Before
running these examples, first runner --add easy-shortcuts to load it into the
static crate, and then runner -C easy-shortcuts to dynamically compile it.
$ runner -xeasy_shortcuts -e 'easy_shortcuts::argn_err(1,"gimme an arg!")' 'an arg'
"an arg"
$ runner -xeasy_shortcuts -e 'easy_shortcuts::argn_err(1,"gimme an arg!")'
/home/steve/.cargo/.runner/bin/tmp error: no argument 1: gimme an arg!
This also applies to --iterator:
$ runner -xeasy_shortcuts -i 'easy_shortcuts::files(".")'
"json.rs"
"print.rs"
With long crate names like this, you can define aliases:
$ runner --alias es=easy_shortcuts
$ runner -xes -e 'es::argn_err(1,"gimme an arg!")'
...
By default, runner -e does a dynamic link, and there are known limitations.
By also using --static, you can evaluate expressions against crates
compiled as static libraries. So, assuming that we have
time in the static cache (runner --add time will do that for you):
$ runner -s -xtime -e "time::now()"
Tm { tm_sec: 34, tm_min: 4, tm_hour: 9, tm_mday: 28, tm_mon: 6, tm_year: 117,
tm_wday: 5, tm_yday: 208, tm_isdst: 0, tm_utcoff: 7200, tm_nsec: 302755857 }
'-X' (--wild) is like -x except it brings all the crate's symbols into scope.
Not something you would overdo in regular code, but it makes for shorter
command lines - the last example becomes (note how short flags can be combined):
$ runner -sXtime -e "now()"
...
-M (--macro) is also like -x except it prepends the 'extern crate' with
#[macro_use]. Consider the very cool r4 crate which
provides list comprehensions. First load in the static cache with runner --add r4,
and then we can say:
$ runner -s --macro r4 -i 'iterate![for x in 0..4; yield x]'
0
1
2
3
Small snippets like these are faster if the crates can be linked dynamically, so
after runner -C r4 to build a shared library in the dynamic cast, you can run this
without the -s.
$ runner -Xto_vec -Mr4 -e 'iterate![for i in 0..2; for j in 0..2; yield (i,j)].to_vec()'
[(0, 0), (0, 1), (1, 0), (1, 1)]
(At this point, the command-line is getting sufficiently complicated that you would be better off with a little snippet that you can edit in a proper editor.)
With -e,-n or -i, you can specify. some initial code with --prepend:
$ runner -p 'let nums=0..5' -i 'nums.clone().zip(nums.skip(1))'
(0, 1)
(1, 2)
(2, 3)
(3, 4)
If you can get away with dynamic linking, then runner can make it
easy to test a module interactively. In this way you get much of the
benefit of a fully interactive interpreter (a REPL):
$ cat universe.rs
pub fn answer() -> i32 {
42
}
$ runner -C universe.rs
building crate 'universe' at universe.rs
$ runner -xuniverse -e "universe::answer()"
42
This provides a way to get to play with big predefined strings:
$ cat > text.rs
pub const TEXT: &str = "possibly very long string";
$ runner -C text.rs
building crate 'text' at text.rs
$ runner -Xtext -e 'TEXT.find("long")'
Some(14)
Consider the example for the filetime crate:
use std::fs;
use filetime::FileTime;
let metadata = fs::metadata("runner.rs").unwrap();
let mtime = FileTime::from_last_modification_time(&metadata);
println!("{}", mtime);
let atime = FileTime::from_last_access_time(&metadata);
assert!(mtime < atime);
// Inspect values that can be interpreted across platforms
println!("{}", mtime.seconds_relative_to_1970());
println!("{}", mtime.nanoseconds());
// Print the platform-specific value of seconds
println!("{}", mtime.seconds());After runner --add filetime, this crate is in your static cache. And runner --doc filetime
will give you its local documentation.
However, it can't be compiled directly, for the reason that use std::fs is already in the runner prelude.
So we need to say:
$ runner -s --no-prelude filetime.rs
1506778536.945440909s
1506778536
945440909
1506778536
Or if you're in a hurry: runner -sN filetime.rs.
As always, can always put these arguments in a first comment like so "//: -sN".
It would be good to provide such an experience for the dynamic-link case, since
it is faster. There is in fact a dynamic cache as well but support for linking
against external crates dynamically is very basic. It works fine for crates that
don't have any external depdendencies, e.g. this creates a libjson.so in the
dynamic cache:
$ runner -C json
And then you can run the json.rs example without -s.
The --compile action takes three kinds of arguments:
- a crate name that is already loaded and known to Cargo
- a Cargo directory
- a Rust source file - the crate name is the file name without extension.
Dynamic linking is not a priority for
Rust tooling at the moment. So we have to build more elaborate libraries without the
help of Cargo. (The following assumes that you have already brought in regex for a Cargo project,
so that the Cargo cache is populated, e.g. with runner --add regex)
runner -C memchr
runner -C aho-corasick
runner -C utf8-ranges
runner -C lazy_static
runner -xlazy_static -C thread_local
runner -C regex-syntax
runner -C regex
This script drives home how tremendously irritating life in Rust would be without Cargo.
We have to track the dependencies, ensure that the correct default features are enabled in the
compilation, and special-case crates which directly link to libc.
However, the results feel worthwhile. Compiling the first regex documented example:
use regex::Regex;
let re = Regex::new(r"^\d{4}-\d{2}-\d{2}$").unwrap();
assert!(re.is_match("2014-01-01"));With a static build (-s) I get 0.56s on this machine, and 0.25s with dynamic linking.
There are limitations to dynamic linking currently - crates which are "no std"
(and don't provide a feature to turn this off) cannot be compiled. Also, remember
that all invocations of runner -C end up with shared libraries placed in one
directory called the 'dynamic cache' - there can only be one crate called 'libs'
for example.