确保这个测试是失败的。这很重要，因为有时你的测试不能重现失败的调用，尤其是你使用了可以混淆测试的 web 或者其他框架。很多东西可能被存储在变量中，但遗憾的是，只通过观察测试，你在测试里调用的东西并不总是明显可见的。当我尝试着重现这个失败的调用时，我不准备说我创建了一个测试，但是，对的，我已经创建了，但我不认为这是特别不寻常的。从自己的错误中吸取教训。
"But Maria," I hear you say, "this is all helpful if I have a stack trace, but I just have a failing test. Where do I start?"
Pdb, the Python Debugger.
Find a place in your code where you know this call should hit. You should be able to find at least one place. Stick a pdb break in there.
#### A digression
Why not a print statement? I used to depend on print statements. They still come in handy sometimes. But once I started working with complicated code bases, and especially ones making network calls, print just became too slow. I ended up with print statements all over the place, I lost track of where they were and why, and it just got complicated. But there is a more important reason to mostly use pdb. Let's say you put a print statement in and discover that something is wrong—and must have gone wrong earlier. But looking at the function where you put the print statement, you have no idea how you got there. Looking at code is a great way to see where you are going, but it is terrible for learning where you've been. And yes, I have done a grep of my code base looking for where a function is called, but this can get tedious and doesn't narrow it down much with a popular function. Pdb can be very helpful.
You follow my advice, and put in a pdb break and run your test. And it whooshes on by and fails again, with no break at all. Leave your breakpoint in, and run a test already in your test suite that does something very similar to the broken test. If you have a decent test suite, you should be able to find a test that is hitting the same code you think your failed test should hit. Run that test, and when it gets to your breakpoint, do a `w` and look at the stack. If you have no idea by looking at the stack how/where the other call may have gone haywire, then go about halfway up the stack, find some code that belongs to you, and put a breakpoint in that file, one line above the one in the stack trace. Try again with the new test. Keep going back and forth, moving up the stack to figure out where your call went off the rails. If you get all the way up to the top of the trace without hitting a breakpoint, then congratulations, you have found the issue: Your app was spelled wrong. No experience here, nope, none at all.
### 4. Change things
If you still feel lost, try making a new test where you vary something slightly. Can you get the new test to work? What is different? What is the same? Try changing something else. Once you have your test, and maybe additional tests in place, it is safe to start changing things in the code to see if you can narrow down the problem. Remember to start troubleshooting with a fresh commit so you can easily back out changes that do not help. (This is a reference to version control, if you aren't using version control, it will change your life. Well, maybe it will just make coding easier. See "[A Visual Guide to Version Control][2]" for a nice introduction.)
### 5. Take a break
In all seriousness, when it stops feeling like a fun challenge or game and starts becoming really frustrating, your best course of action is to walk away from the problem. Take a break. I highly recommend going for a walk and trying to think about something else.
### 6. Write everything down
When you come back, if you aren't suddenly inspired to try something, write down any information you have about the problem. This should include:
* Exactly the call that is causing the problem
* Exactly what happened, including any error messages or related log messages
* Exactly what you were expecting to happen
* What you have done so far to find the problem and any clues that you have discovered while troubleshooting
Sometimes this is a lot of information, but trust me, it is really annoying trying to pry information out of someone piecemeal. Try to be concise, but complete.
### 7. Ask for help
I often find that just writing down all the information triggers a thought about something I have not tried yet. Sometimes, of course, I realize what the problem is immediately after hitting the submit button. At any rate, if you still have not thought of anything after writing everything down, try sending an email to someone. First, try colleagues or other people involved in your project, then move on to project email lists. Don't be afraid to ask for help. Most people are kind and helpful, and I have found that to be especially true in the Python community.
Maria McKinley will present [Hunting the Bugs][3] at [PyCascades 2019][4], February 23-24 in Seattle.
