7 releases
0.2.1 | Feb 3, 2022 |
---|---|
0.2.0 | Feb 2, 2022 |
0.1.4 | Dec 1, 2021 |
0.1.3 | Sep 12, 2021 |
0.1.0 | Mar 10, 2021 |
#56 in Text processing
2,822 downloads per month
Used in tw-econ
56KB
386 lines
sscanf
A Rust crate with a sscanf (inverse of format!()) Macro based on Regex
sscanf
is originally a C-function that takes a String, a format String with placeholders and several
Variables (in the Rust version replaced with Types). It then parses the input String, writing
the values behind the placeholders into the Variables (Rust: returns a Tuple). sscanf
can be
thought of as reversing a call to format!()
:
// format: takes format string and values, returns String
let s = format!("Hello {}{}!", "World", 5);
assert_eq!(s, "Hello World5!");
// scanf: takes String, format string and types, returns Tuple
let parsed = scanf!(s, "Hello {}{}!", str, usize);
// parsed is Result<(&str, usize), sscanf::Error>
assert_eq!(parsed, Ok(("World", 5)));
// alternative syntax: Types in format string
let parsed2 = scanf!(s, "Hello {str}{usize}!");
assert_eq!(parsed, parsed2);
scanf!()
takes a format String like format!()
, but doesn't write
the values into the placeholders ({}
), but extracts the values at those {}
into the return Tuple.
If matching the format string failed, an Error is returned:
let s = "Text that doesn't match the format string";
let parsed = scanf!(s, "Hello {}{}!", str, usize);
assert!(matches!(parsed, sscanf::Error::RegexMatchFailed{..}));
Note that the original C-function and this Crate are called sscanf, which is the technically
correct version in this context. scanf
(with one s
) is a similar C-function that reads a
console input instead of taking a String parameter. The macro itself is called scanf!()
because
that is shorter, can be pronounced without sounding too weird and nobody uses the stdin version
anyway.
Types in Placeholders:
The types can either be given as a separate parameter after the format string, or directly
inside of the {}
placeholder.
The first allows for autocomplete while typing, syntax highlighting and better compiler errors
generated by scanf in case that the wrong types are given.
The second imitates the Rust format!() behavior since 1.58.
This option does not allow any paths (like std::string::String
) or any other form that might
contain a :
, since :
marks the start of Format Options.
It also gives worse compiler errors when using stable Rust.
More examples of the capabilities of scanf
:
let input = "<x=3, y=-6, z=6>";
let parsed = scanf!(input, "<x={i32}, y={i32}, z={i32}>");
assert_eq!(parsed, Ok((3, -6, 6)));
let input = "Move to N36E21";
let parsed = scanf!(input, "Move to {char}{usize}{char}{usize}");
assert_eq!(parsed, Ok(('N', 36, 'E', 21)));
let input = "Escape literal { } as {{ and }}";
let parsed = scanf!(input, "Escape literal {{ }} as {{{{ and }}}}");
assert_eq!(parsed, Ok(()));
let input = "A Sentence with Spaces. Another Sentence.";
// str and String do the same, but String clones from the input string
// to take ownership instead of borrowing.
let (a, b) = scanf!(input, "{String}. {String}.").unwrap();
assert_eq!(a, "A Sentence with Spaces");
assert_eq!(b, "Another Sentence");
// Number format options
let input = "0xab01 0o127 101010 1Z";
let parsed = scanf!(input, "{usize:x} {i32:o} {u8:b} {u32:r36}");
let (a, b, c, d) = parsed.unwrap();
assert_eq!(a, 0xab01); // Hexadecimal
assert_eq!(b, 0o127); // Octal
assert_eq!(c, 0b101010); // Binary
assert_eq!(d, 71); // any radix (r36 = Radix 36)
assert_eq!(d, u32::from_str_radix("1Z", 36).unwrap());
let input = "color: #D4AF37";
// Number types take their size into account, and hexadecimal u8 can
// have at most 2 digits => only possible match is 2 digits each.
let (r, g, b) = scanf!(input, "color: #{u8:x}{u8:x}{u8:x}").unwrap();
assert_eq!((r, g, b), (0xD4, 0xAF, 0x37));
The input here is a &'static str
, but in can be String
, &str
, &String
, ...
Basically anything with AsRef<str>
and without taking Ownership. This also means that the input might need to outlive the
scanf!()
call, because the Error
type borrows from it and using str
returns a slice from the input.
The parsing part of this macro has very few limitations, since it replaces the {}
with a Regular
Expression (regex
) that corresponds to that type.
For example:
char
is just one Character (regex"."
)str
is any sequence of Characters (regex".+?"
)- Numbers are any sequence of digits (regex
"[-+]?\d+"
)
And so on. The actual implementation for numbers tries to take the size of the Type into account and some other details, but that is the gist of the parsing.
This means that any sequence of replacements is possible as long as the Regex finds a
combination that works. In the char, usize, char, usize
example above it manages to assign
the N
and E
to the char
s because they cannot be matched by the usize
s.
Format Options
All Options are inside '{'
'}'
and after a :
. Literal '{'
or '}'
inside of a Format
Option are escaped as '\{'
instead of '{{'
to avoid ambiguity.
Procedural macro don't have any reliable type info and can only compare types by name. This means
that the number options below only work with a literal type like "i32
", NO Paths ()
or Wrappers (std::i32
) or Aliases (struct Wrapper(i32);
). ONLY type Alias = i32;
i32
,
usize
, u16
, ...
config | description | possible types |
---|---|---|
{:/ <regex> /} |
custom regex | any |
{:x} |
hexadecimal numbers | numbers |
{:o} |
octal numbers | numbers |
{:b} |
binary numbers | numbers |
{:r2} - {:r36} |
radix 2 - radix 36 numbers | numbers |
{: <chrono format> } |
chrono format | chrono types |
Custom Regex:
{:/.../}
: Match according to theRegex
between the/
/
For example:
let input = "random Text";
let parsed = scanf!(input, "{str:/[^m]+/}{str}");
// regex [^m]+ matches anything that isn't an 'm'
// => stops at the 'm' in 'random'
assert_eq!(parsed, Ok(("rando", "m Text")));
As mentioned above, '{'
'}'
have to be escaped with a '\'
. This means that:
"{"
or"}"
would give a compiler error"\{"
or"\}"
lead to a"{"
or"}"
inside of the regex- curly brackets have a special meaning in regex as counted repetition
"\\{"
or"\\}"
would give a compiler error- first
'\'
escapes the second one, leaving just the brackets
- first
"\\\{"
or"\\\}"
lead to a"\{"
or"\}"
inside of the regex- the first
'\'
escapes the second one, leading to a literal'\'
in the regex. the third escapes the curly bracket as in the second case - needed in order to have the regex match an actual curly bracket
- the first
Note that this is only the case if you are using raw strings for formats, regular strings require
escaping '\'
, so this would double the number of '\\'
.
Works with non-String
types too:
let input = "Match 4 digits: 123456";
let parsed = scanf!(input, r"Match 4 digits: {usize:/\d\{4\}/}{usize}");
// raw string r"" to write \d instead of \\d
// regex \d{4} matches exactly 4 digits
assert_eq!(parsed, Ok((1234, 56)));
Note that changing the regex of a non-String
type might cause that type's FromStr
to fail
Number Options:
Only work on primitive number types (u8
, ..., u128
, i8
, ..., i128
, usize
, isize
):
x
: hexadecimal Number (Digits 0-9 and A-F or a-f, optional Prefix0x
)o
: octal Number (Digits 0-7, optional Prefix0o
)b
: binary Number (Digits 0-1, optional Prefix0b
)r2
-r36
: any radix Number (no prefix)
chrono
integration (Requires chrono
feature):
The types DateTime
,
NaiveDate
,
NaiveTime
,
NaiveDateTime
,
Utc
and
Local
can be used and accept
a Date/Time format string
inside of the {
}
. This will then be used for both the Regex generation and parsing of the
type.
Using DateTime
returns a
DateTime<FixedOffset>
and requires the rules and limits that DateTime::parse_from_str
has.
use chrono::prelude::*;
let input = "10:37:02";
let parsed = scanf!(input, "{NaiveTime:%H:%M:%S}");
assert_eq!(parsed, Ok(NaiveTime::from_hms(10, 37, 2)));
let expected = Utc.ymd(2020, 5, 23).and_hms(21, 5, 7);
// DateTime<*> directly implements FromStr and doesn't need a config
let input = "2020-05-23T21:05:07Z";
let parsed = scanf!(input, "{DateTime<Utc>}");
assert_eq!(parsed, Ok(expected));
let input = "Today is the 23. of May, 2020 at 09:05 pm and 7 seconds.";
let parsed = scanf!(input, "Today is the {Utc:%d. of %B, %Y at %I:%M %P and %-S} seconds.");
assert_eq!(parsed, Ok(expected));
Note: The chrono
feature needs to be active for this to work, because chrono
is an optional dependency
Custom Types
scanf
works with most of the primitive Types from std
as well as String
by default. The
full list can be seen here: Implementations of RegexRepresentation
.
More Types can easily be added, as long as they implement FromStr
for the parsing
and RegexRepresentation
for scanf
to obtain the Regex of the Type:
struct TimeStamp {
year: usize, month: u8, day: u8,
hour: u8, minute: u8,
}
impl sscanf::RegexRepresentation for TimeStamp {
/// Matches "[year-month-day hour:minute]"
const REGEX: &'static str = r"\[\d\d\d\d-\d\d-\d\d \d\d:\d\d\]";
}
impl std::str::FromStr for TimeStamp {
// ...
}
let input = "[1518-10-08 23:51] Guard #751 begins shift";
let parsed = scanf!(input, "{TimeStamp} Guard #{usize} begins shift");
assert_eq!(parsed, Ok((TimeStamp{
year: 1518, month: 10, day: 8,
hour: 23, minute: 51
}, 751)));
Implementing RegexRepresentation
isn't strictly necessary if you always supply a custom
Regex when using the type by using the {:/.../}
format option, but this tends to make your code
less readable.
Dependencies
~1.6–2.6MB
~64K SLoC