deno/cli/lsp/npm.rs
David Sherret 62e952559f
refactor(permissions): split up Descriptor into Allow, Deny, and Query (#25508)
This makes the permission system more versatile.
2024-09-16 21:39:37 +01:00

131 lines
3.7 KiB
Rust

// Copyright 2018-2024 the Deno authors. All rights reserved. MIT license.
use dashmap::DashMap;
use deno_core::anyhow::anyhow;
use deno_core::error::AnyError;
use deno_core::serde_json;
use deno_semver::package::PackageNv;
use deno_semver::Version;
use serde::Deserialize;
use std::sync::Arc;
use crate::args::npm_registry_url;
use crate::file_fetcher::FileFetcher;
use crate::npm::NpmFetchResolver;
use super::search::PackageSearchApi;
#[derive(Debug)]
pub struct CliNpmSearchApi {
file_fetcher: Arc<FileFetcher>,
resolver: NpmFetchResolver,
search_cache: DashMap<String, Arc<Vec<String>>>,
versions_cache: DashMap<String, Arc<Vec<Version>>>,
}
impl CliNpmSearchApi {
pub fn new(file_fetcher: Arc<FileFetcher>) -> Self {
let resolver = NpmFetchResolver::new(file_fetcher.clone());
Self {
file_fetcher,
resolver,
search_cache: Default::default(),
versions_cache: Default::default(),
}
}
pub fn clear_cache(&self) {
self.file_fetcher.clear_memory_files();
self.search_cache.clear();
self.versions_cache.clear();
}
}
#[async_trait::async_trait]
impl PackageSearchApi for CliNpmSearchApi {
async fn search(&self, query: &str) -> Result<Arc<Vec<String>>, AnyError> {
if let Some(names) = self.search_cache.get(query) {
return Ok(names.clone());
}
let mut search_url = npm_registry_url().join("-/v1/search")?;
search_url
.query_pairs_mut()
.append_pair("text", &format!("{} boost-exact:false", query));
let file_fetcher = self.file_fetcher.clone();
let file = deno_core::unsync::spawn(async move {
file_fetcher
.fetch_bypass_permissions(&search_url)
.await?
.into_text_decoded()
})
.await??;
let names = Arc::new(parse_npm_search_response(&file.source)?);
self.search_cache.insert(query.to_string(), names.clone());
Ok(names)
}
async fn versions(&self, name: &str) -> Result<Arc<Vec<Version>>, AnyError> {
if let Some(versions) = self.versions_cache.get(name) {
return Ok(versions.clone());
}
let info = self
.resolver
.package_info(name)
.await
.ok_or_else(|| anyhow!("npm package info not found: {}", name))?;
let mut versions = info.versions.keys().cloned().collect::<Vec<_>>();
versions.sort();
versions.reverse();
let versions = Arc::new(versions);
self
.versions_cache
.insert(name.to_string(), versions.clone());
Ok(versions)
}
async fn exports(
&self,
_nv: &PackageNv,
) -> Result<Arc<Vec<String>>, AnyError> {
Ok(Default::default())
}
}
fn parse_npm_search_response(source: &str) -> Result<Vec<String>, AnyError> {
#[derive(Debug, Deserialize)]
struct Package {
name: String,
}
#[derive(Debug, Deserialize)]
struct Object {
package: Package,
}
#[derive(Debug, Deserialize)]
struct Response {
objects: Vec<Object>,
}
let objects = serde_json::from_str::<Response>(source)?.objects;
Ok(objects.into_iter().map(|o| o.package.name).collect())
}
#[cfg(test)]
mod tests {
use super::*;
#[test]
fn test_parse_npm_search_response() {
// This is a subset of a realistic response only containing data currently
// used by our parser. It's enough to catch regressions.
let names = parse_npm_search_response(r#"{"objects":[{"package":{"name":"puppeteer"}},{"package":{"name":"puppeteer-core"}},{"package":{"name":"puppeteer-extra-plugin-stealth"}},{"package":{"name":"puppeteer-extra-plugin"}}]}"#).unwrap();
assert_eq!(
names,
vec![
"puppeteer".to_string(),
"puppeteer-core".to_string(),
"puppeteer-extra-plugin-stealth".to_string(),
"puppeteer-extra-plugin".to_string()
]
);
}
}