Fix relative links to README.md files

[chrstnst]

Fixes #1920

I suggest adding to the 'index' preprocessor a call to swap the filenames in links in the chapters' contents.

[chrstnst]

Fixes #984

[sanmai-NL]

@ehuss Is this change easy to review and accept?

[ehuss]

I would prefer to not use regular expressions to translate the links. Can the translation be done in adjust_links or somewhere like that?

[CaydenPierce]

Would love to see this merged.

[sanmai-NL]

@CaydenPierce would you be able to make the requested improvement to the original PR, using a more robust/performant matching algorithm than regular expression matching?

[ghost]

Hi, I've been reading the discussions on this topic and I'm also interested in seeing mdBook incorporate a solution for this issue.

Can the translation be done in adjust_links or somewhere like that?

I had a look and came up with something that extends the existing regular expression in adjust_links. However this would apply to all links, even if the 'index' preprocessor has been disabled. I'm sharing it in case this can help moving forward on this.

diff --git a/src/utils/mod.rs b/src/utils/mod.rs
index 9156916..d0cda17 100644
--- a/src/utils/mod.rs
+++ b/src/utils/mod.rs
@@ -95,7 +95,7 @@ pub fn unique_id_from_content(content: &str, id_counter: &mut HashMap<String, us
 fn adjust_links<'a>(event: Event<'a>, path: Option<&Path>) -> Event<'a> {
     static SCHEME_LINK: Lazy<Regex> = Lazy::new(|| Regex::new(r"^[a-z][a-z0-9+.-]*:").unwrap());
     static MD_LINK: Lazy<Regex> =
-        Lazy::new(|| Regex::new(r"(?P<link>.*)\.md(?P<anchor>#.*)?").unwrap());
+        Lazy::new(|| Regex::new(r"(?P<link>.*?)(?P<readme>README)?\.md(?P<anchor>#.*)?").unwrap());
 
     fn fix<'a>(dest: CowStr<'a>, path: Option<&Path>) -> CowStr<'a> {
         if dest.starts_with('#') {
@@ -126,7 +126,16 @@ fn adjust_links<'a>(event: Event<'a>, path: Option<&Path>) -> Event<'a> {
             }
 
             if let Some(caps) = MD_LINK.captures(&dest) {
-                fixed_link.push_str(&caps["link"]);
+                let link = &caps["link"];
+                fixed_link.push_str(link);
+                // "Links to README.md will be converted to index.html"
+                if let Some(readme) = caps.name("readme") {
+                    if link.is_empty() || link.ends_with('/') {
+                        fixed_link.push_str("index");
+                    } else {
+                        fixed_link.push_str(readme.as_str());
+                    }
+                }
                 fixed_link.push_str(".html");
                 if let Some(anchor) = caps.name("anchor") {
                     fixed_link.push_str(anchor.as_str());

Several questions I can think of:

1. How to deal with the fact that the 'index' preprocessor can be run or not? Maybe the logic should live in src/preprocess/index.rs and make calls to adjust_links.
2. "I would prefer to not use regular expressions to translate the links.": Is it ok to extend the regex in adjust_links?

[bitdivine]

Maybe README.md could be mapped to index.html AND README.html. This way the links don't need to be updated. Just an alternative as changing the links seems to be complicated. I don't believe that having README.md should be a problem for users. After all, users can have either paths ending in / or in /index.html, so why not /README.html as well? It might not look as cool as having URLs end in / but it is simple and robust.

[sanmai-NL]

Maybe README.md could be mapped to index.html AND README.html. This way the links don't need to be updated. Just an alternative as changing the links seems to be complicated. I don't believe that having README.md should be a problem for users. After all, users can have either paths ending in / or in /index.html, so why not /README.html as well? It might not look as cool as having URLs end in / but it is simple and robust.

That would work great, but the canonical address should be recorded in the HTML to prevent duplicate content for indexers like online search engines.

[jlinford]

Looking forward to seeing this fixed. Thanks!