dmaax/ladybird
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

LibWebView: Make the session-history-mirror merge linear

Browse source 
Problem: While a document was concurrently loaded, a burst of sync same-
document history navigations (e.g. a pushState flood) could spin the UI
process at full CPU — and on slow/Sanitizer builds, intermittently time
out other tests (since one WebContent process is reused across tests).

Cause: The UI process keeps an authoritative session-history mirror, and
merges each WebContent snapshot into it. find_merge_anchor compares each
local entry against each incoming one. Every URL comparison serializes
both URLs. When a snapshot briefly diverges from the mirror, the anchor
is no longer near the end. So, the search degraded to a deep quadratic
walk — with a string serialization per-comparison. A flood compounded
that from both ends: Every pushState added a top-level entry — driving
the count each walk must cover into the hundreds — and also triggered a
history update. So, the merge ran again for every one of them.

Fix: Index the incoming entries by serialized URL once — keyed as URL
equality compares (full serialization, fragment included). So, the URL-
keyed anchor searches are linear, not quadratic.
This commit is contained in:
sideshowbarker 2026-06-17 09:40:46 +09:00 committed by Andreas Kling
parent e61a91c017
commit 766bc991ed
1 changed files with 40 additions and 7 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

47
Libraries/LibWebView/SessionHistory.cpp
View file

@ -654,15 +654,48 @@ TraversableSessionHistory::UpdateResult TraversableSessionHistory::update_from_w
});
}
// AD-HOC: The URL-keyed anchor searches are the common case. But find_merge_anchor compares every local entry
// against every incoming entry — and each URL comparison serializes both URLs. A burst of same-document
// navigations (such as a pushState flood) can push the top-level entry count into the hundreds while this
// runs once per history update — so, a quadratic scan with per-comparison serialization becomes a
// UI-process bottleneck. Index the incoming entries by serialized URL once — keyed the same way URL
// equality compares (full serialization, fragment included) — so each URL-keyed search is linear.
if (!merge_anchor.has_value()) {
merge_anchor = find_merge_anchor([](Entry const& local_entry, Entry const& incoming_entry) {
return local_entry.url == incoming_entry.url && entries_have_matching_nonzero_document_state_id(local_entry, incoming_entry);
});
}
if (!merge_anchor.has_value()) {
merge_anchor = find_merge_anchor([](Entry const& local_entry, Entry const& incoming_entry) {
return local_entry.url == incoming_entry.url;
HashMap<String, Vector<size_t>> incoming_indices_by_url;
for (size_t incoming_index = *incoming_current_top_level_entry_index + 1; incoming_index > 0; --incoming_index) {
auto candidate_incoming_index = incoming_index - 1;
incoming_indices_by_url.ensure(entries[candidate_incoming_index].url.serialize()).append(candidate_incoming_index);
}
// The incoming index lists are built highest-first. So, this reproduces find_merge_anchor's choice among the
// local/incoming entry pairs whose URLs are equal and who satisfy the predicate: The highest local entry — and
// for it, the highest matching incoming entry.
auto find_url_merge_anchor = [&](auto predicate) -> Optional<SessionHistoryMergeAnchor> {
for (size_t local_index = *local_current_top_level_entry_index + 1; local_index > 0; --local_index) {
auto candidate_local_index = local_index - 1;
auto incoming_candidates = incoming_indices_by_url.find(m_entries[candidate_local_index].url.serialize());
if (incoming_candidates == incoming_indices_by_url.end())
continue;
for (auto candidate_incoming_index : incoming_candidates->value) {
if (!predicate(m_entries[candidate_local_index], entries[candidate_incoming_index]))
continue;
return SessionHistoryMergeAnchor {
.local_index = candidate_local_index,
.incoming_index = candidate_incoming_index,
};
}
}
return {};
};
merge_anchor = find_url_merge_anchor([](Entry const& local_entry, Entry const& incoming_entry) {
return entries_have_matching_nonzero_document_state_id(local_entry, incoming_entry);
});
if (!merge_anchor.has_value()) {
merge_anchor = find_url_merge_anchor([](Entry const&, Entry const&) {
return true;
});
}
}
if (!merge_anchor.has_value())