Create an also_recognize combinator to make providing source slices fast and safe.

src/parser/combinator.rs

@@ -1,3 +1,5 @@
+use std::ops::RangeTo;
+
 use super::parser_context::ContextElement;
 use super::parser_context::PreviousElementNode;
 use super::token::Token;
@@ -6,6 +8,28 @@ use nom::error::ErrorKind;
 use nom::error::ParseError;
 use nom::IResult;
 use nom::InputLength;
+use nom::Offset;
+use nom::Parser;
+use nom::Slice;
+
+/// Return both the parsed output and the output of recognize() together without having to run the child parser twice.
+pub fn also_recognize<I: Clone + Offset + Slice<RangeTo<usize>>, O, E: ParseError<I>, F>(
+    mut parser: F,
+) -> impl FnMut(I) -> IResult<I, (I, O), E>
+where
+    F: Parser<I, O, E>,
+{
+    move |input: I| {
+        let i = input.clone();
+        match parser.parse(i) {
+            Ok((i, val)) => {
+                let index = input.offset(&i);
+                Ok((i, (input.slice(..index), val)))
+            }
+            Err(e) => Err(e),
+        }
+    }
+}
 
 pub fn context_many1<'r, 's, I, O, E, M>(
     context: Context<'r, 's>,

src/parser/paragraph.rs

@@ -1,3 +1,4 @@
+use super::combinator::also_recognize;
 use super::combinator::context_many_till;
 use super::error::Res;
 use super::parser_context::ChainBehavior;
@@ -26,8 +27,11 @@ pub fn paragraph<'r, 's>(context: Context<'r, 's>, i: &'s str) -> Res<&'s str, P
             exit_matcher: ChainBehavior::AndParent(Some(&context_paragraph_end)),
         }))
         .with_additional_node(ContextElement::StartOfParagraph);
-    let (remaining, (many, till)) =
-        context_many_till(&paragraph_context, text_element, context_paragraph_end)(i)?;
+    let (remaining, (source, (many, till))) = also_recognize(context_many_till(
+        &paragraph_context,
+        text_element,
+        context_paragraph_end,
+    ))(i)?;
     let many = many
         .into_iter()
         .filter_map(|token| match token {
@@ -40,6 +44,7 @@ pub fn paragraph<'r, 's>(context: Context<'r, 's>, i: &'s str) -> Res<&'s str, P
         Paragraph {
             contents: many,
             paragraph_end: till,
+            source,
         },
     ))
 }

src/parser/token.rs

@@ -68,6 +68,7 @@ pub struct Link<'a> {
 
 #[derive(Debug)]
 pub struct Paragraph<'a> {
+    pub source: &'a str,
     pub contents: Vec<TextElement<'a>>,
     pub paragraph_end: &'a str,
 }
@@ -91,22 +92,6 @@ impl<'a> Source<'a> for TextElement<'a> {
 
 impl<'a> Source<'a> for Paragraph<'a> {
     fn get_source(&'a self) -> &'a str {
-        if self.contents.is_empty() {
-            return self.paragraph_end;
-        }
-        // TODO: Is there a better way to do this? At a minimum I should be checking that the pointers are contiguous instead of blindly adding their lengths but maybe theres a good way in nom to get both the recognize() value and the parsed values so we can just store a &str to the source.
-        let start = self.contents[0].get_source().as_ptr();
-        let len = self
-            .contents
-            .iter()
-            .map(|text_element| text_element.get_source().len())
-            .sum::<usize>()
-            + self.paragraph_end.len();
-        let full_source = unsafe {
-            let slice = std::slice::from_raw_parts(start, len);
-            std::str::from_utf8(slice)
-                .expect("A token should always be made with valid utf-8 source material.")
-        };
-        full_source
+        self.source
     }
 }