isomorphic-git 1.41.0 → 1.41.1

package/index.cjs

@@ -8974,21 +8974,76 @@ async function collect(iterable) {
   return result
 }
 
+// The `@` that ends the credentials is the last one in the authority, and the
+// scheme is matched case-insensitively: `HTTPS://user:pass@host/x` is a URL
+// like any other, and leaving the credentials in it is what `fetch` rejects.
+// The authority ends at the first `/`, `?` or `#`, so an `@` that lives in a
+// query or a fragment is left where it is.
+const CREDENTIALS = /^(https?:\/\/)([^/?#]+)@/i;
+
 function extractAuthFromUrl(url) {
   // For whatever reason, the `fetch` API does not convert credentials embedded in the URL
   // into Basic Authentication headers automatically. Instead it throws an error!
   // So we must manually parse the URL, rip out the user:password portion if it is present
   // and compute the Authorization header.
-  // Note: I tried using new URL(url) but that throws a security exception in Edge. :rolleyes:
-  let userpass = url.match(/^https?:\/\/([^/]+)@/);
+  const credentials = url.match(CREDENTIALS);
   // No credentials, return the url unmodified and an empty auth object
-  if (userpass == null) return { url, auth: {} }
-  userpass = userpass[1];
-  const [username, password] = userpass.split(':');
-  // Remove credentials from URL
-  url = url.replace(`${userpass}@`, '');
-  // Has credentials, return the fetch-safe URL and the parsed credentials
-  return { url, auth: { username, password } }
+  if (credentials == null) return { url, auth: {} }
+  // Has credentials, return the fetch-safe URL and the parsed credentials.
+  // The credentials are cut out of the original string rather than read back
+  // off the parsed URL: serializing that one would also drop an explicit
+  // `:443` and append a slash to a URL that has no path, and this return value
+  // is what the request goes out against.
+  return {
+    url: url.replace(CREDENTIALS, '$1'),
+    auth: parseUserinfo(url, credentials[2]),
+  }
+}
+
+function parseUserinfo(url, userpass) {
+  // `URL` splits the two halves the way the platform does, so the first colon
+  // separates them and any later one stays in the password. Only the split is
+  // taken from it, for the reason above.
+  const parsed = parseUrl(url);
+  // A password is absent, not empty, when no colon was written at all. `URL`
+  // reports `''` for both, and the difference is worth keeping.
+  const hasPassword = userpass.indexOf(':') !== -1;
+
+  if (parsed !== null) {
+    return {
+      username: decodeUserinfo(parsed.username),
+      password: hasPassword ? decodeUserinfo(parsed.password) : undefined,
+    }
+  }
+
+  // No `URL` to parse with, or it rejected a string the expression above
+  // accepted. Split on the first colon by hand, without the percent-decoding.
+  const separatorIndex = userpass.indexOf(':');
+  return {
+    username: hasPassword ? userpass.slice(0, separatorIndex) : userpass,
+    password: hasPassword ? userpass.slice(separatorIndex + 1) : undefined,
+  }
+}
+
+function parseUrl(url) {
+  if (typeof URL !== 'function') return null
+  try {
+    return new URL(url)
+  } catch (_) {
+    return null
+  }
+}
+
+// `URL` hands back the userinfo still percent-encoded, so `p:ss` arrives as
+// `p%3Ass` and has to be decoded before it reaches the Authorization header. A
+// lone `%` is not something the parser encodes, and it makes the decode throw,
+// so the raw value stands in whenever the decode fails.
+function decodeUserinfo(value) {
+  try {
+    return decodeURIComponent(value)
+  } catch (_) {
+    return value
+  }
 }
 
 function padHex(b, n) {
@@ -9630,8 +9685,8 @@ function filterCapabilities(server, client) {
 
 const pkg = {
   name: 'isomorphic-git',
-  version: '1.41.0',
-  agent: 'git/isomorphic-git@1.41.0',
+  version: '1.41.1',
+  agent: 'git/isomorphic-git@1.41.1',
 };
 
 class FIFO {

package/index.js

@@ -8961,21 +8961,76 @@ async function collect(iterable) {
   return result
 }
 
+// The `@` that ends the credentials is the last one in the authority, and the
+// scheme is matched case-insensitively: `HTTPS://user:pass@host/x` is a URL
+// like any other, and leaving the credentials in it is what `fetch` rejects.
+// The authority ends at the first `/`, `?` or `#`, so an `@` that lives in a
+// query or a fragment is left where it is.
+const CREDENTIALS = /^(https?:\/\/)([^/?#]+)@/i;
+
 function extractAuthFromUrl(url) {
   // For whatever reason, the `fetch` API does not convert credentials embedded in the URL
   // into Basic Authentication headers automatically. Instead it throws an error!
   // So we must manually parse the URL, rip out the user:password portion if it is present
   // and compute the Authorization header.
-  // Note: I tried using new URL(url) but that throws a security exception in Edge. :rolleyes:
-  let userpass = url.match(/^https?:\/\/([^/]+)@/);
+  const credentials = url.match(CREDENTIALS);
   // No credentials, return the url unmodified and an empty auth object
-  if (userpass == null) return { url, auth: {} }
-  userpass = userpass[1];
-  const [username, password] = userpass.split(':');
-  // Remove credentials from URL
-  url = url.replace(`${userpass}@`, '');
-  // Has credentials, return the fetch-safe URL and the parsed credentials
-  return { url, auth: { username, password } }
+  if (credentials == null) return { url, auth: {} }
+  // Has credentials, return the fetch-safe URL and the parsed credentials.
+  // The credentials are cut out of the original string rather than read back
+  // off the parsed URL: serializing that one would also drop an explicit
+  // `:443` and append a slash to a URL that has no path, and this return value
+  // is what the request goes out against.
+  return {
+    url: url.replace(CREDENTIALS, '$1'),
+    auth: parseUserinfo(url, credentials[2]),
+  }
+}
+
+function parseUserinfo(url, userpass) {
+  // `URL` splits the two halves the way the platform does, so the first colon
+  // separates them and any later one stays in the password. Only the split is
+  // taken from it, for the reason above.
+  const parsed = parseUrl(url);
+  // A password is absent, not empty, when no colon was written at all. `URL`
+  // reports `''` for both, and the difference is worth keeping.
+  const hasPassword = userpass.indexOf(':') !== -1;
+
+  if (parsed !== null) {
+    return {
+      username: decodeUserinfo(parsed.username),
+      password: hasPassword ? decodeUserinfo(parsed.password) : undefined,
+    }
+  }
+
+  // No `URL` to parse with, or it rejected a string the expression above
+  // accepted. Split on the first colon by hand, without the percent-decoding.
+  const separatorIndex = userpass.indexOf(':');
+  return {
+    username: hasPassword ? userpass.slice(0, separatorIndex) : userpass,
+    password: hasPassword ? userpass.slice(separatorIndex + 1) : undefined,
+  }
+}
+
+function parseUrl(url) {
+  if (typeof URL !== 'function') return null
+  try {
+    return new URL(url)
+  } catch (_) {
+    return null
+  }
+}
+
+// `URL` hands back the userinfo still percent-encoded, so `p:ss` arrives as
+// `p%3Ass` and has to be decoded before it reaches the Authorization header. A
+// lone `%` is not something the parser encodes, and it makes the decode throw,
+// so the raw value stands in whenever the decode fails.
+function decodeUserinfo(value) {
+  try {
+    return decodeURIComponent(value)
+  } catch (_) {
+    return value
+  }
 }
 
 function padHex(b, n) {
@@ -9617,8 +9672,8 @@ function filterCapabilities(server, client) {
 
 const pkg = {
   name: 'isomorphic-git',
-  version: '1.41.0',
-  agent: 'git/isomorphic-git@1.41.0',
+  version: '1.41.1',
+  agent: 'git/isomorphic-git@1.41.1',
 };
 
 class FIFO {