Let N = proxy_to_source.size() before the code modified in this commit.
Let M = (N - proxy_start). Let K = source_items.size(). The algorithmic
complexity of the removed loop is O(N+K+K*M), assuming the number of
O(N+K) reallocations is a constant. The complexity of the QList::insert
and std::copy implementation is O(N+K). This is much faster in practice
when K and M are of the same order of magnitude as N.
For example, this quadratic complexity issue results in noticeable
slowdown in the following scenario:
* a QSortFilterProxyModel is used only for filtering, not sorting;
* first set a filter that matches a single item in the middle of a
huge number of items (about one million) - this is reasonably fast
(takes about a second);
* then clear the filter (i.e. set an empty filter so that no item is
filtered out) and watch your application's UI freeze for a minute.
The "Add QSortFilterProxyModel clear-filter benchmark" commit (with
Change-Id I419a5521dd0be7676fbb09b34b4069d4a76423b1) adds a benchmark
that runs much faster with this performance fix.
Pick-to: 6.0 6.1
Change-Id: Ieaec173e6910f5d21eaee49402087f7711abbedf
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
Reviewed-by: Giuseppe D'Angelo <giuseppe.dangelo@kdab.com>
7d92ef63d7