I'm taking it for a spin and sharing my notes as I go:
* UI has pleasant colors, I could get used to comfort of having all relevant info in side panes
* The gdb starts in /opt/gdbfrontend directory, would be nice if it changed to dir you executed gdbfrontent in
* I wish I could collapse Disassembly list - I don't want to see it ever
* Icons don't have tooltips and some are not so clear
* "Sources" pane doesn't actually have all the sources, some appear only after I set breakpoint in previously missing file
* Breakpoints set manually through GDB console don't show up in UI breakpoint list
* Usual shortcuts F10, F11... for stepping over and into don't work, while gdb commands n and s work only if lower pane has focus, so interface requires more mouse clicking
* In my specific case it doesn't detect local variables or respond to added watches (fields to right stay blank)
This is where I stopped using it, don't have time to debug the debugger. I appreciate the effort and it looks like it might be extremely handy tool in future.
That reminds me about that one day in at Sun back in the late 90's when I compiled Solaris in debug mode and set up KGDB.
I was running a full Solaris kernel while anothe rmachine was running GDB and DDD. I was able to visualise all the kernel structures in realtime. It was significantly more usable that I would have imagined it to be.
Can you do something similar these days with Linux?
That's comparable to DTrace. The Solaris kernel debugger, kmdb, is something else, and really worth knowing (if only it was universally adopted by C-coded kernels!).
With kmdb you get to write shell-like pipelines using powerful data structure traversal builtins, and it is safe (you can't crash the system by dereferencing NULL pointers). Really, it's quite fantastic.
curious - the Sun developers I saw in the late 90s had a multi-window debugger but the local variables Did Not Update, and had to be 'refreshed' manually; at the same time, I had MetrwoWerks IDE Debugger on a Macintosh, and easily had all locals, source code, stack, RAM inspection by structure and more, all updating with every change. The Sun devs were being PAID a LOT more, yet their tools looked primitive and the contextual knowledge to use them was much greater..
I'm not sure. I wasn't working as a kernel developer. I did have to debug stuff though to find bugs or explain certain behaviours.
I guess one reason it wasn't used much is because it was a hassle to set up, and I think it got a bit slow when you had a lot of information displayed. I do recall not using this setup much in practice, as just breaking into the debugger on the machine itself was faster and easier.
You had to lot of options when it came to working on the Solaris kernel, and the code was so easy to understand. In fact, I believe the Solaris code base is the best C code I've ever worked with. It's a pity what happened after Oracle took over.
I love the ability of ddd to plot 1d and 2d arrays in gnuplot, while you debug, updating the plot after each next instruction. Really handy for numerical codes. I have not found yet another tool with the same functionality. The python calls from gdb allows to do something similar but does not update automatically and you have to use much more code
What I don't understand with this "modernity" is these "black background shimmery text" themes.
Because that's what we had in bad old days before the monitors were able to show pixels -- the text consoles could actually show only text and the technology used was too poor to have nice pale background and black letters.
But why would somebody want this on the modern displays, except because he does that instead of properly adjusting the background light or because "the real cool guys look at black screens" is beyond me. Seriously, the displays aren't cheap eighties displays anymore. And even then the expensive ones already looked much better by default than these "modern" themes now.
/end of the old guy's lawn rant
Otherwise, I definitely like every debugging tool that tries to have a good GUI.
> What I don't understand with this "modernity" is these "black background shimmery text" themes.
Because we're mostly using emissive displays, and it turns out that trying to imitate what works on paper on an emissive display sucks for long term use (though it looks pretty at a glance, and if you are composing things for print WYSIWYG may be more important than what is most comfortable on the screen.) We tried light themes as the almost invariable rule for the first several decades of GUIs, but recently got over it.
(Light backgrounds are awesome on reflective media, and if eInk was cheap and fast enough to use for primary monitors for coding we’d do that.)
You are really just talking about your preferences as if they are scientific absolutes. It depends on your eyes including the amount of floaters you have and how imperfect the lens of your eye is away from its centre, the lighting conditions of your room, your monitor, your taste.
I am not sure exactly when the recent trend towards dark editor themes started but I wonder if it was around the same time that WLED LCD backlighting took over? I love white themes on my CFL monitor but not so much on my WLED laptop.
I also think maybe many people started preferring dark colour schemes because colours pop and look more saturated against black, and so it looks better in screenshots etc.
> You are really just talking about your preferences as if they are scientific absolutes.
No, I'm talking about what's true of the population as a whole (but not every individual member) as if it was the driver of the trend the poster I was responding to questioned...because it is.
Light themes are better than dark themes from an eye strain perspective if a room is well lit. On the other hand, dark themes are better in low light conditions, so if you like coding in dark rooms dark themes are essential. Dark themes are also good for photo editing where you don’t want chrome saturating content.
> light themes are better than dark themes from an eye strain perspective if a room is well lit.
IIRC, the issue is legibility more than eyestrain (though they are related) and light themes are still better with normal vision, but a majority of the adult population has one or more of the vision issues that reverse this (astigmatism being the main one, and alone accounting for nearly 50%.)
> Because we're mostly using emissive displays, and it turns out that trying to imitate what works on paper on an emissive display sucks for long term use
I claim it’s not true. If one would measure The light flux of my screen I’m sure it wouldn’t be higher than that of the paper when reading in good light conditions. I claim that it’s dependent of how one configures his display.
I used to battle too against "dark modes". But I was forced to use VSCode for one project (since a couple of months), and I have to say that I became so used to it that now I want to use "dark mode" everywhere I can. Unfortunately IDEs for embedded like KEIL uVision will never get such feature, and now it's a pain to go back to white background.
No. It's photons, and the photons count is the same for a given flux. If your screen emits more photons, it's because you've set the background light too high. That's that user configurable setting I've mentioned in the first post. If you have problems with reflection of some other light source and you solved it by turning it off and sitting in the dark room, you of course have to reduce the background light of your monitor to match the ambient light conditions.
Light is light, I know. But reading from e-ink is not as comfortable as reading from an LCD whatever the brightness setting is. There are many reasons, like refresh frequency, angle, etc. I'm sorry I don't have time now, but you can look it up.
"I don’t like this so nobody else should" is always obnoxious to read.
I like dark mode. If it has options to let you have dark-text-on-bright-backgrounds, why does the option for a dark mode bother you?
Tech should cater to different needs. Nobody should have to defend their preferences against anyone, but for those wondering if there are any objective reasons:
I don’t want intense light constantly blasting my eyes, taking up more energy and reducing the display’s effective lifetime, and I want to be able to work comfortably in dark environments, which may be dark for any reason: Not wanting to wake other people up, or bother anyone in theaters/restaurants/cars/planes/etc. or simply reducing energy usage.
I don't think this is true: many transmissive LCD technologies adjust their backlight brightness, including differently across regions, depending on the displayed image?
OLEDs incorporate the backlight with the pixel, so they can turn on or off the light for each pixel, giving you much nicer blacks. A typical LED backlight is just a single uniform light, so it can only be adjusted for the entire screen, and unless your screen is black, you need some light. I guess if everything was a at most a darker grey, they could crank it down, but I don’t think anyone does that?
I'm not sure about typical, but I believe most high-end LED backlit displays use at least one of global or local dimming of the backlight depending on what's on the screen.
This means that the power use of this type of LED backlit display is also sensitive to what is being displayed (but at a coarser resolution to OLED).
It's not "completely fake", it's how your rant reads to a lot of readers - whether you intended it or not. @Razengan was giving you a hand here, helping you understand why that rant is disliked.
I have tried to use a lighter background because apparently "that's the correct thing". After experiencing significant eye fatigue I switched back to dark blue.
The interface is surprisingly good, and it's what I recommend to new gdb users if they aren't using an IDE. I still prefer the Emacs integration[1], but I don't recommend that to people who aren't already experienced Emacs users.
Seconding TUI. Pretty solid. In fact if you need a better debugger than that I see no reason as to not jump to a good IDE ... no other middle ground required.
If you’re looking for something more lightweight in between vanilla GDB and a full GUI frontend, I would highly recommend GDB Dashboard: https://github.com/cyrus-and/gdb-dashboard
It’s just a single file (replaces your ~/.gdbinit) that wraps your GDB session with a nice TUI using the Python API. I’ve found it to be a nice middle ground.
I see it is using PyQt5. Any favorite books/tutorials/guides on learning Qt5 or PyQt5? (I'm at 1/4 of Koenig's accelerated C++ and know some python to get my day to day script needs done)
Is this any different than the debugger visualisations that other editors/IDEs have?
Genuinely curious, because if I'm understanding this correctly, it's just another front-end like what VSCode/Atom(probably)/CLion/Visual Studio have been doing for years if not decades.
I guess what I'm asking is: does this do a better job than the others?
Here is a related video, where Eskill Steenberg shows off his data debugger / live visualisation. It's networked so can be used remotely. Very impressive stuff.
Not having active development isn't necessarily a good reason to avoid something, since it still works perfectly well.. There are only so many features needed in a debugger and security is not a real concern.
Would be amazing if this could also integrate with a language server to provide code browsing functionality. These days I'm using Vim's gdb integration, which doesn't have any data visualisation features, but lets me use all of my code browsing plugins.
visual studio code is pretty good these days when you get the right extensions. I've setup my projects to be debugged with gdb with vscode as the GUI. My build system also generate a compile_commands.json to feed to clangd https://marketplace.visualstudio.com/items?itemName=llvm-vs-...
Another good project that people love to hate is eclipse, if you want a GDB powered GUI, its always worth a shot.
With all of the improvements to static and run-time analysis that we see within compilers like -fsanitize, will it ever get to the point were tools like GDB become obsolete? It seems to me that as more work goes into this, it will become very hard for gdb/valgrind/... to compete with the full power/knowledge available to a compiler.
