An unwinder tries to take over the work from gdb of unwinding the stack, which can be quite a messy thing and sometimes go wrong
In a first pass we let the gdb unwinder do its thing and just cache the results (cache gets invalidated when the program stops again etc.). Then when unwinding is done next, we can jump in. For that we have to invalidate the register cache of gdb so it calls us again, which has some nasty side effects too, which we try to remedy.
During that we can chose to return the values extracted from gdb, or we can replace some values for some registers in certain stack frames. This may lead to leaving out stack frames, causing gdb to find one it didn't find before, or to recover parameters and registers in a different way.
Since we often need the frame before or after it, a lot of things cannot be done for the last or first frame.
Since we don't want to do all the debug information handling, we leave it to gdb to do most of the heavy lifting, which means that often we need to cache all the unwinding and then only in a second pass can do our stuff. For this we sometimes invalidate the register cache which can have ugly side effects.
Also we are often wrong about registers, so you need to be careful about it.
Best is to leave it disabled and only use it for difficult scenarios where you know what you are doing.
Enables or disables on the fly. Will cause certain caches to be flushed.
This tries to completely hide from gdb the existence of a frame by pretending the next after it can
Causes the register in the specified frame to be returned differently. All other functionality is basically implemented using this primitive.
vdb-unwind-enable
The default state of the unwinder. It is recommended to leave it disabled until needed.
Here you can see how after frame 7 gdb gets utterly confused and spits out nonsense.
After manually telling it that frame 8 is really a different position, gdb is able to get further (until again it loses
track. We could go on there for a next round if we so wanted)
This command will help you search the stack for return pointers that originated from a function call to the current frame. It will mark the most likely call, but otherwise output all valid code fragments that things on the stack point to.
You can change its behaviour by the following configurations
vdb-unwind-hint-marker
Is the text that gets displayed when the hint heuristics think this is a very likely match.
vdb-unwind-colors-hint
This is the color the marker gets colored in.
vdb-unwind-hint-context
This is the context string given to the dis/ command for disassembling the context around the possible match.
vdb-unwind-colors-hint-start
This is how the highlighting of a new hint match start message is done.
For the above example of the frame 7, we do the unwind hint there. As you can see the heuristics tell us about the
computed jump possibly being a good path. No others are marked. However with a bit of experience on knows that certain
calls (especially those that may never return) are internally replaced by direct jumps, so the call to __cxa_throw is
very likely the one that is happening there, and so we can try it and it makes the backtrace a little more correct.