[Const-me]

> Figuring out how you got there isn't necessarily any easier with a debugger than with generous logging. If anything, it can be much harder.

Put a breakpoint early on startup, once hit put a data breakpoint on the part of the state which messed up, reproduce the bug, and there's very high chance debugger will take you to the code which broke the state.

This will even happen if the code which breaks the state is a memory corruption bug in different thread, in the code written in another language, from a third-party DLL.

> debuggers are nice for simple, borderline trivial programs where everything fits on a screen and there's a handful of variables to keep track of

It's funny I think it's the opposite. When there's only a few variables, one can print/log the complete state pretty often. When the state takes a gigabyte of memory and changes often, similar amount of logging going to produce too many terabytes of logs to be useful. Debugging is interactive, you can inspect the complete state and find the most relevant pieces to watch.

[foxfluff]

> Put a breakpoint early on startup, once hit put a data breakpoint on the part of the state which messed up, reproduce the bug, and there's very high chance debugger will take you to the code which broke the state.

Except when you find the data you're after doesn't exist early at startup, it's allocated on the fly as connections come and go, and you need thousands of connections to come and go before the bug manifests, and the data that eventually gets messed up may first be used thousands of times before it's wrong all of a sudden. IME it's precisely these multi-threaded memory corruption bugs that really resist debuggers. And as is often the case, debugger changes timing so much that the bug doesn't even reproduce.

Even if you find the code that corrupts the memory, it might be totally okay, it just somehow got passed the wrong memory long ago through no fault of its own (or you're looking at a use-after-free).

[Const-me]

Indeed, it can happen too. All software is different, bugs are different, and optimal debugging tactics is very different as well.

But when the one in my comment does work, the saved debugging time can be measured in days.

Speaking about debugging tactics, there’re way more than just two. There’re OS-level tools like process monitor on Windows / strace on Linux. There’re specialized tools like RenderDoc or Wireshark. For different types of bugs, these tools can sometimes also save days of work compared to other methods.

[foxfluff]

Aye. It's also not always obvious what's the optimal tactic, and I believe in many cases there are multiple equally valid approaches and you pick the one you feel most comfortable with. Sometimes the one you start with is a dead end.

That's why I don't really like an absolutist stance (if you rarely use debugger you're dumb / if you always poke around with a debugger you're dumb). I just happen to fall in the "rarely uses debugger" camp myself.

> I’ve done hundreds of interviews like this, and it’s fascinating watching what people do. Do they read the code first? Fire up a debugger? Add print statements and binary search by hand? [..] After hundreds of interviews I still couldn’t tell you which approach is best.

(From https://news.ycombinator.com/item?id=29813036)

[gjulianm]

> Put a breakpoint early on startup, once hit put a data breakpoint on the part of the state which messed up, reproduce the bug, and there's very high chance debugger will take you to the code which broke the state.

Knowing which part of the state messed up is like 90% of the debugging. Most of the time the part of the state you see messed up is like that because another part of the state is also messed up. Given that neither debuggers nor logging can go back in time to backtrack the state (well, time travel debuggers exist but aren't common or cover all cases), it's usually faster to inspect the code manually to try to find the original bug instead of re-running continuously to backtrack the state.

[Const-me]

> it's usually faster to inspect the code manually

I’m not sure how usual that is. Sometimes “the code” is too many thousands of lines to inspect. Other times, “the code” is only the machine code but not the source code, inspecting megabytes of disassembly is not fun.

[gjulianm]

If you're tracking changes to a given state and seeing which other parts of the state affected those changes, you end up inspecting the same spots, whether it's with a debugger or not.

> inspecting megabytes of disassembly is not fun.

Well, of course there you need the debugger and memory watchers. But I think that fits into the "unusual" slot, most people aren't doing that.