Hacker News Reader

Comment by jrockway | Hacker News Reader

Most people that use vim also don't use graphical debuggers. I use the "print" statement.

Spewing print statements all over a piece of code is as bad as any other shotgun debugging technique. The reason breakpoints and watches were invented were to reduce reliance on bad ways of doing things.

What's good about print statements is that they don't require me to do anything more than once. When you run code in a debugger, there is a lot of manual interaction involved. continue, break, see where you are, think about that, poke around, repeat. With print statements, you run your test, analyze the output, and make the changes. It's all automated except for the analysis step, which is all the human should be doing.

If you need breakpoints, watchpoints, and a GUI to click around in, your code is too complicated and unmaintainable, which is probably why you need a debugger in the first place.

Manual interaction? Sorry, but I'm still very skeptical. Pausing in a debugger does not require materially more "manual interaction" than endlessly running your program to spit out some printed statements--it just front-loads the manual interaction so you're doing it sooner and, if you're doing it right, doing less of it. A debugger allows live autopsy, which is far more valuable than this silly print-statement thing--when your print statement does not return as expected, you are not much better off with your print statements than just seeing the broken output of the program.

Couple live autopsy with live edit-and-continue and you've got a recipe for a considerably more effective way of debugging code.

And "your code is too complicated and unmaintainable if you need a debugger" is preposterous on its face if you've ever...oh, I don't know...written a driver? Or any other remotely complex (you're conflating complicated and complex) task with a significant number of interlocking parts? (Hell, one could just as easily say that your code is too complicated and unmaintainable if you're barfing print statements all over the place.)

You "step over" instead of "step into". Now what? You have to start over.

With test cases and print statements, starting over is free. If you need more information, edit the code to give it to you and re-run.

Getting a debugged program back to the desired state is, in my experience, difficult even with an automatic test. Not only does the program state have to be right, but so does the debugger state. Without the debugger, you have half the state to bother with.

I like logging for this, but the concept is basically the same. Being able to switch on debug logging on a running system has saved the day more than once for me.

Thing is, logging and debugging are not in opposition. My current web app is...well, verbose is probably putting it mildly. But when developing locally, if something goes sideways I just say "hey, stop here" and I can introspect the entire kit and kaboodle without even having to re-bounce the debug server (and certainly without stopping it to bung in a bunch of print statements).

I agree somewhat (although I do feel that debuggers are useful) in that IDEs can influence the design of your program by encouraging a debug first development style similar to "programming by coincidence"

Nobody is putting print statements all over the place. At least I don't do. I set it where I want to inspect values. If you want a separate piece of software for that, more power to you, but please don't tell me my nails must be crooked because I use a different hammer than you.

This assumes you know ahead of time exactly what values must be inspected to solve the problem. This is frequently very much not the case, especially if you're ever working on somebody else's code.

So, yes, your nails may be straight, but your chosen (regressive) technique leaves you with no way to know if they're straight and with only the option of pounding harder and harder if the nail doesn't go in the first time. Enjoy!

Of course I know, that's why I read the code for. Understanding the code base is my job. Using print statements to exclude factors which could contribute to the problem is not different from using a debugger to do the same.

If you debug without knowing what you're debugging and just looking for getting it running, it sounds like you're the one doing the shotgun-programming/debugging, separate debugger or not.

I'm not sure what code you write or what it's for, but in the real world, sometimes we deal with buggy, archaic and downright bizarre code that we didn't write, and sometimes we have deadlines.

So, yes. I use a debugger, so that I can spend some time at home too. Because as nice as the office is, it's summertime, and there's beer in my fridge.

Back when I was seventeen and staying up all night writing a video game engine though, I definitely would have been onboard with the whole print debugging idea.

Nice ad hominem, doesn't work though. I'm 30 and have been developing for 14 years now. And yes, it's real world. And it's mostly Perl, so I do know about "buggy, archaic and downright bizarre code." I'm still more productive with print statements than having a separate system poke around in the applications.

But again, I must be wrong because I'm doing it different, right? So the only option can be that I'm too young, don't know what I'm talking about, or just lack experience in general, right? There couldn't be a possible way that someone else uses a different solution that works just as fine for them, right?

Considering that I never said "Don't use a debugger." only that "I don't need a debugger", you guys seem pretty worked up considering that you're claiming you're the more relaxed one.

It's a tool, don't take it personally that I have a different method that works. It doesn't invalidate yours. At least not in my view.

Honest question (as I avoid Perl like the plague) - is there an introspective debugger on par with Visual Studio's in Perl-land?

Because if you're leaning on print statements in absence of appropriate tooling, I can understand that. When the tooling is available, however, I am skeptical of claims as to enhanced productivity or any other positivity coming from it.

There's a way to get a good answer from someone, compare his language of choice to a deadly disease. And then ask if it's as good as some proprietary crapware.

What does visual studio? Show the value of variables in separate windows as the code runs? Yes, the perl debugger can do that. You can also inject code and poke the entire program state.

The perl debugger is a perl function that is called for every line and every subroutine entry. Use that to build whatever tool you want, and good luck finding that in Visual Studio.

"Proprietary crapware." Yeah, that's the sign of a really well-balanced, competent developer.

I never said I have enhanced productivity over using a debugger. I said I have no worse productivity.

There's a debugger, there's a REPL, and in general lots of tools which I use if I see the need (the NYTProf profiler would be one). Using a debugger in this scenario (with an interpreted language) just doesn't give me any enhanced productivity over what I'm doing now.

Someday, you'll get to work in a language that has a good REPL, and you'll never want to go back to printing. :)

I'm sorry, but this is condescending and simply false, even with the smiley face. I use output for inspection in Perl, Scheme, Javascript, and for (what are in my world) good reasons. The print statement usually doesn't influence the running program in any way. That means no matter who wrote the interpreter/compiler, the debugger, or whatnot, the printing won't have much influence on the program. But I've heard "Why is this bug only showing up in the debugger?" more than once in my time.

Someday, you'll want to run your code more than once, and you'll write tests instead of typing them into the REPL.

Print statements aren't meaningful tests, though. Hence the eight hojillion xUnit ports for pretty much every major language.

OK, so say you have some test that creates a foobar object and invokes the method baz, which does some stuff that you care about. The test fails. What do you do? Do you pick apart the implementation of baz in the REPL, or do you add some print statements inside baz?

The second option is easier. You obviously remove them once your understanding (and tweaking) of the code leads to a passing test.

you set a trace point inside baz, investigate with the REPL, and fix your code.

wait.. what? are you scanning stdout with your tests?

The only downside to print statements is forgetting to remove them and having some obnoxious message committed.

A useful trick we've had at previous workplaces which solves this (and related accidental-debug-code-checkin) problems is to put a precommit hook that denies a checkin with a certain string in the text (we used "DNCI" for Do Not Check In)

So you get in the habit of annotating test tweaks & debug prints with:

printf("DNCI x is %d\n", x);

I use gdb as well, but sometimes dumping data directly is all you really need.

This is rarely a problem when using tools like darcs, magit, or git-add --patch.

Reply on news.ycombinator.com