Is there a way to make Python stop on error when running PyStata ?

[John Williamss]

Suppose I have the following (toy) code

import stata_setup
stata_setup.config("C:/Program Files/Stata17", "mp")

for i in range(1000):
 %stata replace y=2

and I run the whole thing in python. Suppose Stata throws an error that variable y does not exist. If it were the real Stata, this would stop the program, and I would be happy. But in python, it will be executed 1000 times! Is there any way to stop it on its own?

[Daniel Schaefer]

What software are you using to edit and run your python script?

[John Williamss]

What software are you using to edit and run your python script?

Spyder but I don't think it matters.

[Daniel Schaefer]

It does matter. It is the responsibility of the IDE to stop on an exception. Just one example of many; pycharm will not stop on an exception when you press "run", but it will stop on an exception when you press "debug". This is useful in software engineering because it allows you to test compiled code without symbolic linking before you release the product. Python is compiled to bycode, so this is a feature (not a bug) in python.

Start by wrapping your example code (minus the loop) in a try except statement and print the type of exception when you hit the except block to the console. This will confirm that an exception is thrown by the Stata "magic" command. Next, check to make sure the IDE is properly configured and that you are running the script in debug mode. If the magic command throws an exception at all, I guarantee the IDE is misconfigured.

[John Williamss]

It does matter. It is the responsibility of the IDE to stop on an exception. Just one example of many; pycharm will not stop on an exception when you press "run", but it will stop on an exception when you press "debug". This is useful in software engineering because it allows you to test compiled code without symbolic linking before you release the product. Python is compiled to bycode, so this is a feature (not a bug) in python.

Start by wrapping your example code (minus the loop) in a try except statement and print the type of exception when you hit the except block to the console. This will confirm that an exception is thrown by the Stata "magic" command. Next, check to make sure the IDE is properly configured and that you are running the script in debug mode. If the magic command throws an exception at all, I guarantee the IDE is misconfigured.

I wrapped according to your instruction and %stata does not throw any error. On the other hand, stata.run throws an error. This is pretty limiting as %stata is more convenient.

[Daniel Schaefer]

Sorry to hear that. It doesn't look like there is any reason why IPython should prevent a command like this from raising an exception, so this was apparently an implementation choice. You might be able to cut this gordian's knot by using the multi-command %%stata magic and writing the for loop in Stata native.

[John Williamss]

Sorry to hear that. It doesn't look like there is any reason why IPython should prevent a command like this from raising an exception, so this was apparently an implementation choice. You might be able to cut this gordian's knot by using the multi-command %%stata magic and writing the for loop in Stata native.

Unfortunately, the only way is by stata.run() . the magic %%stata does not work in a python loop.

[Daniel Schaefer]

I am suggesting that, rather than wrapping the line in the python loop, you might be able to write the loop you need in Stata's ado language using the multiline magic command. Of course, there may be cases where that isn't appropriate. It depends on what, exactly, you are trying to do.

[Tim Huegerich]

It may help to understand this puzzling %stata behavior to know that pystata.run used to also not raise an error (contrary to its documentation) until a bug fix in November of last year.

[John Williamss]

It may help to understand this puzzling %stata behavior to know that pystata.run used to also not raise an error (contrary to its documentation) until a bug fix in November of last year.

Pretty interesting. I should probably report the bug there.

[Tim Huegerich]

Unfortunately, the only way is by stata.run() . the magic %%stata does not work in a python loop.

As an aside, the `sfi` package provides another way to run single-line commands, a way which can be much faster than `stata.run` (or `%stata`) in the looping scenario you describe: sfi.SFIToolkit.stata()

I'm not sure whether this is widely known. For details (and some caveats), see: https://hugetim.github.io/nbstata/stata.html#run_single