This Bugzilla instance is a read-only archive of historic NetBeans bug reports. To report a bug in NetBeans please follow the project's instructions for reporting issues.
Created attachment 146318 [details] C++ source I am trying to convince my group to use smart pointers instead of regular C pointers. The main objection from the developers is that it is harder to see values in the debugger. OK, I made a simple experiment, I created a tree using shared_ptr and using regular pointers (we need shared_ptr because nodes can have dup pointers). Then I ran debugger and expanded the tree in Variables window. The tree has three levels. With plain pointers I need to expand three levels in the variable value tree to reach the NULL pointer. With smart pointers I need to expand 11 levels. First view takes 7 lines, second view requires 15 lines. I am attaching two snapshots from the debugger. I am also attaching source code for simplicity. Value balloon is also much longer with a little reason. BTW, many problems in C++ messages are related with inability to convert compiler expanded types back to user's types. I wrote such back converters in C++ and Python and know that it is mostly doable. It may require some heuristic but it really worth to do. Refcount is so useless in this context. It is a pointer to something containing a real counter. Do I need to know implementation details? I prefer to see two fields in shared_ptr: pointer and counter, plain, no wrappers. We need to switch to smart pointers but I have a weak position in ease of debugging. Please help. Product Version: NetBeans IDE 8.0 Beta (Build 201401141042) Java: 1.7.0_51; Java HotSpot(TM) Client VM 24.51-b03 Runtime: Java(TM) SE Runtime Environment 1.7.0_51-b13 A hint: why does every node waste a line with no text and then next line has one more hierarchy level with a type (std::__shared_ptr<...>)? Why not combine them into one line? P.S. Yes, I know, you can say that there is a nice script that I can source in GDB/NetBeans which can allow to make C++/STL human friendlier in debugging. Is it right way? These days nobody reads long manuals. If every user needs to setup and then support some init scripts, the tool cannot be called user friendly out of the box.
Created attachment 146319 [details] Two snapshots of NB 8
Eugene, if I understand correctly, such things are usually done through pretty printers. And IDE allows to specify gdb ini file where pretty printers are initialized. I.e. you can find pretty printers for STL in Internet.
I had to learn how to use/create pretty printers but with one of STL printer implementation I see this in my gdb console: (gdb) p root std::shared_ptr (count 1, weak 0) 0x605028 without printer I see (gdb) p root $4 = {<std::__shared_ptr<Node, (__gnu_cxx::_Lock_policy)2>> = {_M_ptr = 0x605028, _M_refcount = {_M_pi = 0x605010}}, <No data fields>} Looking at Python code I suspect that full type is stripped from typename argument of the printer and it is impossible to improve what we see above. As the result, in Variables tab, NB shows a plain pointer value without an ability to expand the hierarchy. So, without printers I need many clicks to see the tree. With printers I cannot see the tree. Where am I wrong?