503 lines
17 KiB
Rust
503 lines
17 KiB
Rust
/// A page, can be a blog post or a basic page
|
|
use std::collections::HashMap;
|
|
use std::path::{Path, PathBuf};
|
|
|
|
use tera::{Tera, Context as TeraContext};
|
|
use slug::slugify;
|
|
use slotmap::{Key};
|
|
|
|
use errors::{Result, ResultExt};
|
|
use config::Config;
|
|
use utils::fs::{read_file, find_related_assets};
|
|
use utils::site::get_reading_analytics;
|
|
use utils::templates::render_template;
|
|
use front_matter::{PageFrontMatter, InsertAnchor, split_page_content};
|
|
use rendering::{RenderContext, Header, render_content};
|
|
use library::Library;
|
|
|
|
use content::file_info::FileInfo;
|
|
use content::ser::SerializingPage;
|
|
|
|
|
|
#[derive(Clone, Debug, PartialEq)]
|
|
pub struct Page {
|
|
/// All info about the actual file
|
|
pub file: FileInfo,
|
|
/// The front matter meta-data
|
|
pub meta: PageFrontMatter,
|
|
/// The list of parent sections
|
|
pub ancestors: Vec<Key>,
|
|
/// The actual content of the page, in markdown
|
|
pub raw_content: String,
|
|
/// All the non-md files we found next to the .md file
|
|
pub assets: Vec<PathBuf>,
|
|
/// All the non-md files we found next to the .md file as string for use in templates
|
|
pub serialized_assets: Vec<String>,
|
|
/// The HTML rendered of the page
|
|
pub content: String,
|
|
/// The slug of that page.
|
|
/// First tries to find the slug in the meta and defaults to filename otherwise
|
|
pub slug: String,
|
|
/// The URL path of the page
|
|
pub path: String,
|
|
/// The components of the path of the page
|
|
pub components: Vec<String>,
|
|
/// The full URL for that page
|
|
pub permalink: String,
|
|
/// The summary for the article, defaults to None
|
|
/// When <!-- more --> is found in the text, will take the content up to that part
|
|
/// as summary
|
|
pub summary: Option<String>,
|
|
/// The earlier page, for pages sorted by date
|
|
pub earlier: Option<Key>,
|
|
/// The later page, for pages sorted by date
|
|
pub later: Option<Key>,
|
|
/// The lighter page, for pages sorted by weight
|
|
pub lighter: Option<Key>,
|
|
/// The heavier page, for pages sorted by weight
|
|
pub heavier: Option<Key>,
|
|
/// Toc made from the headers of the markdown file
|
|
pub toc: Vec<Header>,
|
|
/// How many words in the raw content
|
|
pub word_count: Option<usize>,
|
|
/// How long would it take to read the raw content.
|
|
/// See `get_reading_analytics` on how it is calculated
|
|
pub reading_time: Option<usize>,
|
|
}
|
|
|
|
|
|
impl Page {
|
|
pub fn new<P: AsRef<Path>>(file_path: P, meta: PageFrontMatter) -> Page {
|
|
let file_path = file_path.as_ref();
|
|
|
|
Page {
|
|
file: FileInfo::new_page(file_path),
|
|
meta,
|
|
ancestors: vec![],
|
|
raw_content: "".to_string(),
|
|
assets: vec![],
|
|
serialized_assets: vec![],
|
|
content: "".to_string(),
|
|
slug: "".to_string(),
|
|
path: "".to_string(),
|
|
components: vec![],
|
|
permalink: "".to_string(),
|
|
summary: None,
|
|
earlier: None,
|
|
later: None,
|
|
lighter: None,
|
|
heavier: None,
|
|
toc: vec![],
|
|
word_count: None,
|
|
reading_time: None,
|
|
}
|
|
}
|
|
|
|
pub fn is_draft(&self) -> bool {
|
|
self.meta.draft
|
|
}
|
|
|
|
/// Parse a page given the content of the .md file
|
|
/// Files without front matter or with invalid front matter are considered
|
|
/// erroneous
|
|
pub fn parse(file_path: &Path, content: &str, config: &Config) -> Result<Page> {
|
|
let (meta, content) = split_page_content(file_path, content)?;
|
|
let mut page = Page::new(file_path, meta);
|
|
page.raw_content = content;
|
|
let (word_count, reading_time) = get_reading_analytics(&page.raw_content);
|
|
page.word_count = Some(word_count);
|
|
page.reading_time = Some(reading_time);
|
|
page.slug = {
|
|
if let Some(ref slug) = page.meta.slug {
|
|
slug.trim().to_string()
|
|
} else if page.file.name == "index" {
|
|
if let Some(parent) = page.file.path.parent() {
|
|
slugify(parent.file_name().unwrap().to_str().unwrap())
|
|
} else {
|
|
slugify(page.file.name.clone())
|
|
}
|
|
} else {
|
|
slugify(page.file.name.clone())
|
|
}
|
|
};
|
|
|
|
if let Some(ref p) = page.meta.path {
|
|
page.path = p.trim().trim_left_matches('/').to_string();
|
|
} else {
|
|
page.path = if page.file.components.is_empty() {
|
|
page.slug.clone()
|
|
} else {
|
|
format!("{}/{}", page.file.components.join("/"), page.slug)
|
|
};
|
|
}
|
|
if !page.path.ends_with('/') {
|
|
page.path = format!("{}/", page.path);
|
|
}
|
|
|
|
page.components = page.path.split('/')
|
|
.map(|p| p.to_string())
|
|
.filter(|p| !p.is_empty())
|
|
.collect::<Vec<_>>();
|
|
page.permalink = config.make_permalink(&page.path);
|
|
|
|
Ok(page)
|
|
}
|
|
|
|
/// Read and parse a .md file into a Page struct
|
|
pub fn from_file<P: AsRef<Path>>(path: P, config: &Config) -> Result<Page> {
|
|
let path = path.as_ref();
|
|
let content = read_file(path)?;
|
|
let mut page = Page::parse(path, &content, config)?;
|
|
|
|
if page.file.name == "index" {
|
|
let parent_dir = path.parent().unwrap();
|
|
let assets = find_related_assets(parent_dir);
|
|
|
|
if let Some(ref globset) = config.ignored_content_globset {
|
|
// `find_related_assets` only scans the immediate directory (it is not recursive) so our
|
|
// filtering only needs to work against the file_name component, not the full suffix. If
|
|
// `find_related_assets` was changed to also return files in subdirectories, we could
|
|
// use `PathBuf.strip_prefix` to remove the parent directory and then glob-filter
|
|
// against the remaining path. Note that the current behaviour effectively means that
|
|
// the `ignored_content` setting in the config file is limited to single-file glob
|
|
// patterns (no "**" patterns).
|
|
page.assets = assets.into_iter()
|
|
.filter(|path|
|
|
match path.file_name() {
|
|
None => true,
|
|
Some(file) => !globset.is_match(file)
|
|
}
|
|
).collect();
|
|
} else {
|
|
page.assets = assets;
|
|
}
|
|
|
|
page.serialized_assets = page.serialize_assets();
|
|
} else {
|
|
page.assets = vec![];
|
|
}
|
|
|
|
Ok(page)
|
|
}
|
|
|
|
/// We need access to all pages url to render links relative to content
|
|
/// so that can't happen at the same time as parsing
|
|
pub fn render_markdown(
|
|
&mut self,
|
|
permalinks: &HashMap<String, String>,
|
|
tera: &Tera,
|
|
config: &Config,
|
|
anchor_insert: InsertAnchor,
|
|
) -> Result<()> {
|
|
let mut context = RenderContext::new(
|
|
tera,
|
|
config,
|
|
&self.permalink,
|
|
permalinks,
|
|
anchor_insert,
|
|
);
|
|
|
|
context.tera_context.insert("page", &SerializingPage::from_page_basic(self, None));
|
|
|
|
let res = render_content(&self.raw_content, &context)
|
|
.chain_err(|| format!("Failed to render content of {}", self.file.path.display()))?;
|
|
|
|
self.summary = res.summary_len.map(|l| res.body[0..l].to_owned());
|
|
self.content = res.body;
|
|
self.toc = res.toc;
|
|
|
|
Ok(())
|
|
}
|
|
|
|
/// Renders the page using the default layout, unless specified in front-matter
|
|
pub fn render_html(&self, tera: &Tera, config: &Config, library: &Library) -> Result<String> {
|
|
let tpl_name = match self.meta.template {
|
|
Some(ref l) => l.to_string(),
|
|
None => "page.html".to_string()
|
|
};
|
|
|
|
let mut context = TeraContext::new();
|
|
context.insert("config", config);
|
|
context.insert("current_url", &self.permalink);
|
|
context.insert("current_path", &self.path);
|
|
context.insert("page", &self.to_serialized(library));
|
|
|
|
render_template(&tpl_name, tera, &context, &config.theme)
|
|
.chain_err(|| format!("Failed to render page '{}'", self.file.path.display()))
|
|
}
|
|
|
|
/// Creates a vectors of asset URLs.
|
|
fn serialize_assets(&self) -> Vec<String> {
|
|
self.assets.iter()
|
|
.filter_map(|asset| asset.file_name())
|
|
.filter_map(|filename| filename.to_str())
|
|
.map(|filename| self.path.clone() + filename)
|
|
.collect()
|
|
}
|
|
|
|
pub fn to_serialized<'a>(&'a self, library: &'a Library) -> SerializingPage<'a> {
|
|
SerializingPage::from_page(self, library)
|
|
}
|
|
|
|
pub fn to_serialized_basic<'a>(&'a self, library: &'a Library) -> SerializingPage<'a> {
|
|
SerializingPage::from_page_basic(self, Some(library))
|
|
}
|
|
}
|
|
|
|
impl Default for Page {
|
|
fn default() -> Page {
|
|
Page {
|
|
file: FileInfo::default(),
|
|
meta: PageFrontMatter::default(),
|
|
ancestors: vec![],
|
|
raw_content: "".to_string(),
|
|
assets: vec![],
|
|
serialized_assets: vec![],
|
|
content: "".to_string(),
|
|
slug: "".to_string(),
|
|
path: "".to_string(),
|
|
components: vec![],
|
|
permalink: "".to_string(),
|
|
summary: None,
|
|
earlier: None,
|
|
later: None,
|
|
lighter: None,
|
|
heavier: None,
|
|
toc: vec![],
|
|
word_count: None,
|
|
reading_time: None,
|
|
}
|
|
}
|
|
}
|
|
|
|
#[cfg(test)]
|
|
mod tests {
|
|
use std::collections::HashMap;
|
|
use std::io::Write;
|
|
use std::fs::{File, create_dir};
|
|
use std::path::Path;
|
|
|
|
use tera::Tera;
|
|
use tempfile::tempdir;
|
|
use globset::{Glob, GlobSetBuilder};
|
|
|
|
use config::Config;
|
|
use super::Page;
|
|
use front_matter::InsertAnchor;
|
|
|
|
|
|
#[test]
|
|
fn test_can_parse_a_valid_page() {
|
|
let content = r#"
|
|
+++
|
|
title = "Hello"
|
|
description = "hey there"
|
|
slug = "hello-world"
|
|
+++
|
|
Hello world"#;
|
|
let res = Page::parse(Path::new("post.md"), content, &Config::default());
|
|
assert!(res.is_ok());
|
|
let mut page = res.unwrap();
|
|
page.render_markdown(
|
|
&HashMap::default(),
|
|
&Tera::default(),
|
|
&Config::default(),
|
|
InsertAnchor::None,
|
|
).unwrap();
|
|
|
|
assert_eq!(page.meta.title.unwrap(), "Hello".to_string());
|
|
assert_eq!(page.meta.slug.unwrap(), "hello-world".to_string());
|
|
assert_eq!(page.raw_content, "Hello world".to_string());
|
|
assert_eq!(page.content, "<p>Hello world</p>\n".to_string());
|
|
}
|
|
|
|
#[test]
|
|
fn test_can_make_url_from_sections_and_slug() {
|
|
let content = r#"
|
|
+++
|
|
slug = "hello-world"
|
|
+++
|
|
Hello world"#;
|
|
let mut conf = Config::default();
|
|
conf.base_url = "http://hello.com/".to_string();
|
|
let res = Page::parse(Path::new("content/posts/intro/start.md"), content, &conf);
|
|
assert!(res.is_ok());
|
|
let page = res.unwrap();
|
|
assert_eq!(page.path, "posts/intro/hello-world/");
|
|
assert_eq!(page.components, vec!["posts", "intro", "hello-world"]);
|
|
assert_eq!(page.permalink, "http://hello.com/posts/intro/hello-world/");
|
|
}
|
|
|
|
#[test]
|
|
fn can_make_url_from_slug_only() {
|
|
let content = r#"
|
|
+++
|
|
slug = "hello-world"
|
|
+++
|
|
Hello world"#;
|
|
let config = Config::default();
|
|
let res = Page::parse(Path::new("start.md"), content, &config);
|
|
assert!(res.is_ok());
|
|
let page = res.unwrap();
|
|
assert_eq!(page.path, "hello-world/");
|
|
assert_eq!(page.components, vec!["hello-world"]);
|
|
assert_eq!(page.permalink, config.make_permalink("hello-world"));
|
|
}
|
|
|
|
#[test]
|
|
fn can_make_url_from_path() {
|
|
let content = r#"
|
|
+++
|
|
path = "hello-world"
|
|
+++
|
|
Hello world"#;
|
|
let config = Config::default();
|
|
let res = Page::parse(Path::new("content/posts/intro/start.md"), content, &config);
|
|
assert!(res.is_ok());
|
|
let page = res.unwrap();
|
|
assert_eq!(page.path, "hello-world/");
|
|
assert_eq!(page.components, vec!["hello-world"]);
|
|
assert_eq!(page.permalink, config.make_permalink("hello-world"));
|
|
}
|
|
|
|
#[test]
|
|
fn can_make_url_from_path_starting_slash() {
|
|
let content = r#"
|
|
+++
|
|
path = "/hello-world"
|
|
+++
|
|
Hello world"#;
|
|
let config = Config::default();
|
|
let res = Page::parse(Path::new("content/posts/intro/start.md"), content, &config);
|
|
assert!(res.is_ok());
|
|
let page = res.unwrap();
|
|
assert_eq!(page.path, "hello-world/");
|
|
assert_eq!(page.permalink, config.make_permalink("hello-world"));
|
|
}
|
|
|
|
#[test]
|
|
fn errors_on_invalid_front_matter_format() {
|
|
// missing starting +++
|
|
let content = r#"
|
|
title = "Hello"
|
|
description = "hey there"
|
|
slug = "hello-world"
|
|
+++
|
|
Hello world"#;
|
|
let res = Page::parse(Path::new("start.md"), content, &Config::default());
|
|
assert!(res.is_err());
|
|
}
|
|
|
|
#[test]
|
|
fn can_make_slug_from_non_slug_filename() {
|
|
let config = Config::default();
|
|
let res = Page::parse(Path::new(" file with space.md"), "+++\n+++", &config);
|
|
assert!(res.is_ok());
|
|
let page = res.unwrap();
|
|
assert_eq!(page.slug, "file-with-space");
|
|
assert_eq!(page.permalink, config.make_permalink(&page.slug));
|
|
}
|
|
|
|
#[test]
|
|
fn can_specify_summary() {
|
|
let config = Config::default();
|
|
let content = r#"
|
|
+++
|
|
+++
|
|
Hello world
|
|
<!-- more -->"#.to_string();
|
|
let res = Page::parse(Path::new("hello.md"), &content, &config);
|
|
assert!(res.is_ok());
|
|
let mut page = res.unwrap();
|
|
page.render_markdown(
|
|
&HashMap::default(),
|
|
&Tera::default(),
|
|
&config,
|
|
InsertAnchor::None,
|
|
).unwrap();
|
|
assert_eq!(page.summary, Some("<p>Hello world</p>\n".to_string()));
|
|
}
|
|
|
|
#[test]
|
|
fn page_with_assets_gets_right_info() {
|
|
let tmp_dir = tempdir().expect("create temp dir");
|
|
let path = tmp_dir.path();
|
|
create_dir(&path.join("content")).expect("create content temp dir");
|
|
create_dir(&path.join("content").join("posts")).expect("create posts temp dir");
|
|
let nested_path = path.join("content").join("posts").join("with-assets");
|
|
create_dir(&nested_path).expect("create nested temp dir");
|
|
let mut f = File::create(nested_path.join("index.md")).unwrap();
|
|
f.write_all(b"+++\n+++\n").unwrap();
|
|
File::create(nested_path.join("example.js")).unwrap();
|
|
File::create(nested_path.join("graph.jpg")).unwrap();
|
|
File::create(nested_path.join("fail.png")).unwrap();
|
|
|
|
let res = Page::from_file(
|
|
nested_path.join("index.md").as_path(),
|
|
&Config::default(),
|
|
);
|
|
assert!(res.is_ok());
|
|
let page = res.unwrap();
|
|
assert_eq!(page.file.parent, path.join("content").join("posts"));
|
|
assert_eq!(page.slug, "with-assets");
|
|
assert_eq!(page.assets.len(), 3);
|
|
assert_eq!(page.permalink, "http://a-website.com/posts/with-assets/");
|
|
}
|
|
|
|
#[test]
|
|
fn page_with_assets_and_slug_overrides_path() {
|
|
let tmp_dir = tempdir().expect("create temp dir");
|
|
let path = tmp_dir.path();
|
|
create_dir(&path.join("content")).expect("create content temp dir");
|
|
create_dir(&path.join("content").join("posts")).expect("create posts temp dir");
|
|
let nested_path = path.join("content").join("posts").join("with-assets");
|
|
create_dir(&nested_path).expect("create nested temp dir");
|
|
let mut f = File::create(nested_path.join("index.md")).unwrap();
|
|
f.write_all(b"+++\nslug=\"hey\"\n+++\n").unwrap();
|
|
File::create(nested_path.join("example.js")).unwrap();
|
|
File::create(nested_path.join("graph.jpg")).unwrap();
|
|
File::create(nested_path.join("fail.png")).unwrap();
|
|
|
|
let res = Page::from_file(
|
|
nested_path.join("index.md").as_path(),
|
|
&Config::default(),
|
|
);
|
|
assert!(res.is_ok());
|
|
let page = res.unwrap();
|
|
assert_eq!(page.file.parent, path.join("content").join("posts"));
|
|
assert_eq!(page.slug, "hey");
|
|
assert_eq!(page.assets.len(), 3);
|
|
assert_eq!(page.permalink, "http://a-website.com/posts/hey/");
|
|
}
|
|
|
|
#[test]
|
|
fn page_with_ignored_assets_filters_out_correct_files() {
|
|
let tmp_dir = tempdir().expect("create temp dir");
|
|
let path = tmp_dir.path();
|
|
create_dir(&path.join("content")).expect("create content temp dir");
|
|
create_dir(&path.join("content").join("posts")).expect("create posts temp dir");
|
|
let nested_path = path.join("content").join("posts").join("with-assets");
|
|
create_dir(&nested_path).expect("create nested temp dir");
|
|
let mut f = File::create(nested_path.join("index.md")).unwrap();
|
|
f.write_all(b"+++\nslug=\"hey\"\n+++\n").unwrap();
|
|
File::create(nested_path.join("example.js")).unwrap();
|
|
File::create(nested_path.join("graph.jpg")).unwrap();
|
|
File::create(nested_path.join("fail.png")).unwrap();
|
|
|
|
let mut gsb = GlobSetBuilder::new();
|
|
gsb.add(Glob::new("*.{js,png}").unwrap());
|
|
let mut config = Config::default();
|
|
config.ignored_content_globset = Some(gsb.build().unwrap());
|
|
|
|
let res = Page::from_file(
|
|
nested_path.join("index.md").as_path(),
|
|
&config,
|
|
);
|
|
|
|
assert!(res.is_ok());
|
|
let page = res.unwrap();
|
|
assert_eq!(page.assets.len(), 1);
|
|
assert_eq!(page.assets[0].file_name().unwrap().to_str(), Some("graph.jpg"));
|
|
}
|
|
}
|