Fixes an issue on Python3 with our stderr redirecting.

[pscadding]

I'm unsure if this is the right fix or not.

I understand that in Python 2, doing

se = open("/my_file.txt", "a+", 0)
se.write("hi")
import time
time.sleep(15)

Would mean that the "hi" text would be written to the file before the script ended without a flush because we provided the 0 unbuffered argument.
However in Python 3 this is only available for binary objects. So you can use the following instead:

se = open("/my_file.txt", "a+b", 0)

This would cause an error with the original script because we are passing a string, but if we did:

import os, sys

stderr = "/test.txt"
se = open(stderr, "a+b", 0)
os.dup2(se.fileno(), sys.stderr.fileno())

a = 8/0

That actually is fine and the error is written to the file. I'm unsure why that is OK when an error is generated, I presume it is being converted to a binary?
But also I'm not sure of the importance here of setting the buffer to 0 on an error, presumably it will be flushed as soon as it errors since the program will stop? And even if it didn't then is it important since we are from what I can tell discarding it anyway?

So perhaps this would be fine instead?

se = open("/my_file.txt", "a+")

[pscadding]

I just spotted the link formatting was broken here.

[eshokrgozar]

Makes sense to me

[jfboismenu]

I have a question. Any idea why we need to do add binary mode to stderr but not stdout? They both point to a file, right, so wouldn't the same issue be present?

[pscadding]

Yeah I don't know, it's why I posed the question in the PR description. I'm not sure on the importance of the binary mode on the stderr, as it appears to work without it. I erred on the side of not changing it.

[jfboismenu]

I don't understand what you mean by I'm not sure on the importance of the binary mode on the stderr, as it appears to work without it. I erred on the side of not changing it.

Your PR added the b next to a+, so you did change it. It is the only change. If you removed it there would be no diff beside the comments.

[pscadding]

sorry, what I meant was that I could have removed the 0 and it would have worked, but I opted to not change that behaviour, and simply preserve that behaviour in Python 3 by adding the b.

[jfboismenu]

stderr on Python is unbuffered, which means that everything you sent to it is sent out immediately, while stdout is buffered and will only write out something to the screen or a file when the content is too big or a newline is found.

There's a Stackoverflow (obviously) post out this: https://stackoverflow.com/questions/19990589/stderr-and-stdout-not-buffered-vs-buffered

[jfboismenu]

Obviously, the post is about C, but I'm assuming this holds for Python as well, since it's written in C.

[pscadding]

yeah I'd read that as well, and understood that it would cause it to be output without needing to be flushed, but in practice it didn't seem to make any difference to the outcome, since on error it would end the script and presumably flush? At least, I didn't appear to have any issues with getting the error output without the binary mode.