Beginners and experts (Batchelder blog post)

[Chris Angelico]

On Sat, Sep 30, 2017 at 2:42 AM, Steve D'Aprano
<steve+python at pearwood.info> wrote:
> Oh, and I'd like to make a (moderate) defense of a kind of "bug fixing by random
> perturbation". Obviously making unrelated, arbitrary changes to code is bad.
> But making non-arbitrary but not fully understood changes to relevant code
> sections can be useful in (at least) two scenarios.
>
> (1) I know there's a bug in a specific chunk of code, but I'm having trouble
> working out where. When everything else fails, if I perturb the code a bit
> (reorder lines, calculate things in a different order, rename variables, etc)
> it may change the nature of the bug enough for me to understand what's
> happening.
>
> That's not *random* or *arbitrary* changes, but they are changes not directed at
> any specific outcome other than "make the code a bit different, and see if the
> error changes". I'd like to say it is the debugging technique of last resort,
> except its perhaps not quite as *last* resort as I'd like, especially in code
> I'm not familiar with.
>
> Its an experiment, but not really "carefully designed". Its more like "what
> happens if we hit this bit with a hammer?" except that programmers, unlike
> engineers, have the luxury of an Undo switch :-)

Sometimes, when I'm debugging something with one of my students, I'll
say something like "Let's do something stupid". That prefaces a
suggested change that is, perhaps:
* logging something that, by all sane logic, cannot possibly be wrong;
* altering the form of a piece of code in a way that shouldn't affect anything;
* changing something that logically should break the code worse, not fix it;
* or worse.

They're still not "random" changes, but when you exhaust all the
logical and sane things to try, sometimes you do something stupid and
it reveals the bug. I wouldn't EVER tell someone to assume that
they've hit a language or library bug - but if you make a meaningless
change and now it works, maybe that's what you've hit. It does happen.

"Why does my program crash when it talks to THIS server, but it's fine
talking to THAT server?"

... half an hour of debugging later ...

"Okay, so THIS server supports elliptic curve cryptography, but THAT
one doesn't. Great. That still doesn't explain the crash."

... two hours of debugging later ...

"Huh. Maybe this library has a bug with ECC?"

That's pretty close to what happened to me today, with the exception
that I spent less time on it (because I had the library's source code
handy). But in any case, it's a good reason to occasionally try
something stupid.

ChrisA