search

EpicGamesExt / raddebugger

A native, user-mode, multi-process, graphical debugger.
MIT License
3.42k stars 174 forks source link

~6 bugs in one short video: Modules, Registers showing bogus output, cannot edit watch expressions, watch expression view rule not respected, assembly view not showing interactable registers, watch address location not correct #337

Open seanptmaher opened 5 hours ago

seanptmaher commented 5 hours ago

Debugger version 9.12 ALPHA - Oct 14 2024 [cc2848f] (most recent release on github)

I am debugging a component debug build of Chromium. I can provide you with more info if you'd like.

As the title (which may get clipped) spells out: ~6 bugs in one short video:

Here is the video in question:

When there is a "bug," I tend to circle my mouse around it :P

https://drive.google.com/file/d/1bcjgRgQ1t_0uvXYzWZEL501NHgCGE3fV/

ryanfleury commented 4 hours ago

Without testing Chromium locally, I believe I've fixed the registers/modules view problems as of 1865c337041d296a2174f92e4ebcb96e8795d2e8.

watch expression view rule not respected, it shows it in hex with the enum name.

This should be fixed as of 53c624a27cee0416f35c4dc44f2044541500f7ea.

To try out the above fixes, you can pull the repository, switch to the dev branch, and build locally (you can get a release copy - which I expect will be much desire for Chromium - via build raddbg release). It should be pretty easy to build locally if you haven't already, the instructions are in the repository readme if you need them.

As for the others:

screen flashes red, doesn't give any reason why

The reason is in the bottom right-hand corner of the window, but it likely just says that committing failed. I'll try to recreate that one locally, I have a suspicion of what is causing that.

assembly view not showing interactable registers, only those AFTER $rip

I think this is likely just an assembly syntax lexing problem (it looks like the lexer got confused at some point) - to get around it temporarily, you can select the register name and hover it.

watch address location not correct, as shown in my jumping to that address and it containing garbage.

This is not a bug, as the watch pin "address location" is referring to the spot in disassembly where you've attached that pin, not the evaluated address of the expression. (You can see how that address changes if you drag/drop that pin around). When you navigate to the address of the pin, that address is the location within the disassembly that you've attached that pin to. So when you navigate there and it shows the various byte values, those are the byte values of the machine code, not of the expression you're evaluating.

ryanfleury commented 4 hours ago

Okay, that watch window mutation bug should also be fixed as of b3d6b2913e925c74701700a10b3b855ada2da0cb. You should be able to write to that value.

You can't modify in the memory view yet, but this is just a known limitation of the memory view; that UI will be more fleshed out a bit later on.

seanptmaher commented 4 hours ago

Thanks for the super fast response!

seanptmaher commented 3 hours ago

In the interest of not creating more issues:

Here's a new video showing stepping being broken when it has to actually step over a function call:

https://drive.google.com/file/d/1EGY2o3tFKJRtaSW8eUXDgVOgz37O2Hbn

Basically, it freezes, "running" the target, but it's actually paused at the correct 'ret' instr.

Then, pausing, and stepping once more will return to the right location.

seanptmaher commented 3 hours ago

Thanks for your work on this by the way, you're doing god's work.....

seanptmaher commented 3 hours ago

If you'd rather I create new issues for these minute little things for easier bugtracking, let me know and I'll do that.

Hovering over a stack frame with extensive template use, the types of the parameters cause the name to be super long, pushing the entire signature out of screen. Can that be changed so it's left-justified and won't go wider than the width of my screen, either wrapping or pushing off to the right?

image