From 3593c3b687a7a67ace5787aa5e9a8a1b22a5bfcf Mon Sep 17 00:00:00 2001
From: Andreas Kling <andreas@ladybird.org>
Date: Thu, 23 Oct 2025 21:45:00 +0200
Subject: [PATCH] LibWeb: Throw out decoded UTF-32 data in HTMLTokenizer after
 parser runs

This ends up saving quite a bit of memory on many pages, since UTF-32
uses 4 bytes per code points.

As an example, it reduces the footprint on https://gymgrossisten.com/
by 2 MiB.
---
 Libraries/LibWeb/HTML/Parser/HTMLParser.cpp    |  2 ++
 Libraries/LibWeb/HTML/Parser/HTMLTokenizer.cpp | 14 ++++++++++++++
 Libraries/LibWeb/HTML/Parser/HTMLTokenizer.h   |  2 ++
 3 files changed, 18 insertions(+)

diff --git a/Libraries/LibWeb/HTML/Parser/HTMLParser.cpp b/Libraries/LibWeb/HTML/Parser/HTMLParser.cpp
index 94a5b86fbfa..d4043934049 100644
--- a/Libraries/LibWeb/HTML/Parser/HTMLParser.cpp
+++ b/Libraries/LibWeb/HTML/Parser/HTMLParser.cpp
@@ -259,6 +259,8 @@ void HTMLParser::run(HTMLTokenizer::StopAtInsertionPoint stop_at_insertion_point
     }
 
     flush_character_insertions();
+
+    m_tokenizer.parser_did_run({});
 }
 
 void HTMLParser::run(URL::URL const& url, HTMLTokenizer::StopAtInsertionPoint stop_at_insertion_point)
diff --git a/Libraries/LibWeb/HTML/Parser/HTMLTokenizer.cpp b/Libraries/LibWeb/HTML/Parser/HTMLTokenizer.cpp
index e61985512d8..ccf08bfa9df 100644
--- a/Libraries/LibWeb/HTML/Parser/HTMLTokenizer.cpp
+++ b/Libraries/LibWeb/HTML/Parser/HTMLTokenizer.cpp
@@ -2895,6 +2895,20 @@ HTMLTokenizer::HTMLTokenizer(StringView input, ByteString const& encoding)
     m_source_positions.empend(0u, 0u);
 }
 
+void HTMLTokenizer::parser_did_run(Badge<HTMLParser>)
+{
+    // OPTIMIZATION: If we've consumed all input and the insertion point is at the start,
+    //               we can throw away the decoded input buffer to save memory.
+    if (m_current_offset > 0
+        && static_cast<size_t>(m_current_offset) == m_decoded_input.size()
+        && (!m_insertion_point.has_value() || *m_insertion_point == 0)
+        && (!m_old_insertion_point.has_value() || *m_old_insertion_point == 0)) {
+        m_decoded_input.clear();
+        m_current_offset = 0;
+        m_prev_offset = 0;
+    }
+}
+
 void HTMLTokenizer::insert_input_at_insertion_point(StringView input)
 {
     Vector<u32> new_decoded_input;
diff --git a/Libraries/LibWeb/HTML/Parser/HTMLTokenizer.h b/Libraries/LibWeb/HTML/Parser/HTMLTokenizer.h
index 7c1c827da95..fc664d5417f 100644
--- a/Libraries/LibWeb/HTML/Parser/HTMLTokenizer.h
+++ b/Libraries/LibWeb/HTML/Parser/HTMLTokenizer.h
@@ -145,6 +145,8 @@ public:
     // This permanently cuts off the tokenizer input stream.
     void abort() { m_aborted = true; }
 
+    void parser_did_run(Badge<HTMLParser>);
+
 private:
     void skip(size_t count);
     Optional<u32> next_code_point(StopAtInsertionPoint);