From 5c36410b54b1b5b21e150c415490ff78f592f79a Mon Sep 17 00:00:00 2001
From: Han Xiao <han.xiao@jina.ai>
Date: Mon, 17 Mar 2025 14:23:02 +0800
Subject: [PATCH] fix: normalize url

---
 src/agent.ts           | 11 ++++++-----
 src/utils/url-tools.ts | 42 +++++++++++++++++++++++++++++++++++++-----
 2 files changed, 43 insertions(+), 10 deletions(-)

diff --git a/src/agent.ts b/src/agent.ts
index 4c96249..0c062ee 100644
--- a/src/agent.ts
+++ b/src/agent.ts
@@ -403,6 +403,7 @@ export async function getResponse(question?: string,
 
   const allURLs: Record<string, SearchSnippet> = {};
   const visitedURLs: string[] = [];
+  const badURLs: string[] = [];
   const evaluationMetrics: Record<string, EvaluationType[]> = {};
   // reserve the 10% final budget for the beast mode
   const regularBudget = tokenBudget * 0.9;
@@ -515,14 +516,13 @@ export async function getResponse(question?: string,
           allKnowledge,
           allURLs,
           visitedURLs,
+          badURLs,
           SchemaGen,
           currentQuestion
         );
 
-        // is this really required???
-        // if (!evaluationMetrics[currentQuestion].includes('attribution')) {
-        //   evaluationMetrics[currentQuestion].push('attribution')
-        // }
+        // remove references whose urls are in badURLs
+        thisStep.references = thisStep.references.filter(ref => !badURLs.includes(ref.url));
       }
 
       updateContext({
@@ -764,6 +764,7 @@ You decided to think out of the box or cut from a completely different angle.
           allKnowledge,
           allURLs,
           visitedURLs,
+          badURLs,
           SchemaGen,
           currentQuestion
         );
@@ -908,7 +909,7 @@ But unfortunately, you failed to solve the issue. You need to think out of the b
     result: thisStep,
     context,
     visitedURLs: returnedURLs,
-    readURLs: visitedURLs,
+    readURLs: visitedURLs.filter(url => !badURLs.includes(url)),
     allURLs: weightedURLs.map(r => r.url)
   };
 }
diff --git a/src/utils/url-tools.ts b/src/utils/url-tools.ts
index 90ca47e..b6078db 100644
--- a/src/utils/url-tools.ts
+++ b/src/utils/url-tools.ts
@@ -390,14 +390,17 @@ export async function processURLs(
   allKnowledge: KnowledgeItem[],
   allURLs: Record<string, SearchSnippet>,
   visitedURLs: string[],
+  badURLs: string[],
   schemaGen: Schemas,
   question: string
-): Promise<{ urlResults: any[], success: boolean }> {
+): Promise<{ urlResults: any[], success: boolean, badURLs: string[] }> {
   // Skip if no URLs to process
   if (urls.length === 0) {
-    return {urlResults: [], success: false};
+    return {urlResults: [], success: false, badURLs: []};
   }
 
+  const badHostnames: string[] = [];
+
   // Track the reading action
   const thisStep: VisitAction = {
     action: 'visit',
@@ -455,12 +458,29 @@ export async function processURLs(
         });
 
         return {url, result: response};
-      } catch (error) {
+      } catch (error: any) {
         console.error('Error reading URL:', url, error);
+        badURLs.push(url);
+        // Extract hostname from the URL
+        if (
+          (error?.name === 'ParamValidationError' && error.message?.includes('Domain')) ||
+          (error?.name === 'AssertionFailureError' && error.message?.includes('resolve host name')) ||
+          error?.message?.includes("Couldn't resolve host name") ||
+          error?.message?.includes("could not be resolved")
+        ) {
+          let hostname = '';
+          try {
+            hostname = extractUrlParts(url).hostname;
+          } catch (e) {
+            console.error('Error parsing URL for hostname:', url, e);
+          }
+          badHostnames.push(hostname);
+          console.log(`Added ${hostname} to bad hostnames list`);
+        }
         return null;
       } finally {
         // Only add valid URLs to visitedURLs list
-        if (url && typeof url === 'string') {
+        if (url) {
           visitedURLs.push(url);
         }
       }
@@ -470,8 +490,20 @@ export async function processURLs(
   // Filter out null results without changing the original array
   const validResults = urlResults.filter(Boolean);
 
+  // remove any URL with bad hostnames from allURLs
+  if (badHostnames.length > 0) {
+    Object.keys(allURLs).forEach(url => {
+        if (badHostnames.includes(extractUrlParts(url).hostname)) {
+          delete allURLs[url];
+          console.log(`Removed ${url} from allURLs`);
+        }
+      }
+    )
+  }
+
   return {
     urlResults: validResults,
-    success: validResults.length > 0
+    success: validResults.length > 0,
+    badURLs
   };
 }
\ No newline at end of file