Try using GStringChunks in PacketListRecord.

This saves a fair amount of memory in tests here. Loading a large
capture file and sorting on a custom column (tcp.window_size) uses 676
MB before the change and 634 after.

Add notes about possble further improvements:

Roll our own replacement for GStringChunks using wmem_tree.

Have PacketListRecord::columnString return a const char * instead of a
const QByteArray.

Change-Id: Icb36194f5ad290828d7106ccc3bf494d07d76d08
Reviewed-on: https://code.wireshark.org/review/10476
Reviewed-by: Gerald Combs <gerald@wireshark.org>

1 files changed, 7 insertions, 1 deletions

diff --git a/ui/qt/packet_list_record.h b/ui/qt/packet_list_record.h
index c8d2e90a28..7c23c43ea8 100644
--- a/ui/qt/packet_list_record.h
+++ b/ui/qt/packet_list_record.h
@@ -36,6 +36,7 @@
 #include <QVariant>
 
 struct conversation;
+struct _GStringChunk;
 
 class PacketListRecord
 {
@@ -55,9 +56,11 @@ public:
     inline int lineCount() { return lines_; }
     inline int lineCountChanged() { return line_count_changed_; }
 
+    static void clearStringPool();
+
 private:
     /** The column text for some columns */
-    QList<QByteArray> col_text_;
+    QList<const char *> col_text_;
 
     frame_data *fdata_;
     int lines_;
@@ -75,6 +78,9 @@ private:
 
     void dissect(capture_file *cap_file, bool dissect_color = false);
     void cacheColumnStrings(column_info *cinfo);
+
+    static struct _GStringChunk *string_pool_;
+
 };
 
 #endif // PACKET_LIST_RECORD_H