[gui] Improve Code Search Performance by Avoiding Exact Node Resolution

[eybisi]

Issue details

Issue

Jadx currently finds the exact node location for code searches. This significantly slows down search performance when retrieving a large number of results, especially for common text searches.

Optimization Proposal

When opening a new tab in the search results, getEnclosingNode is used to create a CodeNode in the search dialog. However, since the item's position is already known, resolving the exact node location is unnecessary. Instead, we can use the class node of the found text directly:

Current code (slow search):

JNode enclosingNode = getOrElse(getEnclosingNode(javaClass, end), rootCls);

Proposed change (instant search):

JNode enclosingNode = convert(wrapper.getJavaNodeByRef(javaClass.getCodeNodeRef()));

Results

With this change, search performance improves significantly(instant). I have tested this modification, clicking on search results correctly jumps to the expected location in Jadx without the additional overhead of exact node resolution.

Jadx version

dev

Java version

21.0.5

OS

• Windows
• Linux
• macOS

[eybisi]

For the reference

Slow search:

Fast search:

[skylot]

If found line is inside method, enclosing node will be that method and Node column will set to that method, same for inner classes. But after your change, enclosing node will always be set to top class. Sure, since in your case metadata for resolving enclosing node is loaded from disk it will be slow. On the other hand, it feels pointless to use Load all button for so many results (200K+).
It is possible to improve enclosing node search by storing node definition positions in separate small files to speed up load, but I am not sure it if really worth it.

As a result: your change indeed speeds up search, but disabling enclosing node resolving give less information about found line and feels like regression.