BF: interpreter EOF format is not specified

[Voileexperiments]

https://esolangs.org/wiki/Brainfuck#Implementation_issues

EOF is a controversial issue. Many implementations return 0, some return -1, and several notable brainfuck programmers suggest that on EOF the memory cell should be left as-is, without modification. In the original distribution the compiler left the cell unchanged, while the interpreter used the EOF from C which, strictly speaking, could be any negative integer, but which usually is -1 (which, in the case of byte-sized cells, ends up as 255).

The current interpreter used for the BF language version uses 0xFF as EOF, which poses a trap for both kata makers and kata solvers:

• input needs to be zero-terminated manually in the input, or they really aren't terminated at all
• we cannot assume that reading pass the end of the input will zero a cell, so we need to do [-] instead of , to zero a cell.

At least this should be documented somewhere.

[kazk]

Is this what you mean?

• For the program input, when EOF is reached, note that a value of -1 is stored in the cell under the pointer, which is considered a standard implementation, albeit not the most popular one. Keep that in mind when writing your BF programs as programs that assume otherwise is likely to cause an infinite loop.

https://github.com/Codewars/codewars-runner-cli/blob/master/documentation/environments/esolangs.md

/cc @DonaldKellett

---

Older docs in documentation/environments needs to be reviewed and reflected to codewars.com wiki so it's easier to find.

[Voileexperiments]

Yes.

I haven't encountered that piece of information anywhere (for as long as I've been doing BF katas). I'm not quite sure where to put this, since it should not be put in example tests because it's an important implementation detail, but putting in on the wiki does not help either (nobody should read the runner environment docs to know the implementation details before they attempt any katas. That page is mostly for reference on setting up tests).

[Bubbler-4]

We have this wiki page, but it's not maintained very well since it has been created. I think we can fill in the necessary language specifics there (e.g. the , command writes 255 to the current cell on EOF).

The pages also need the full list of currently supported extra packages (e.g. Python has numpy, tensorflow, etc... but it's not listed there; it's listed here in this repo's doc but it's rather hard to find IMO; Dockerfile shows even more packages, but no clue what they are for.) along with links to their own documentations whenever possible. (Package without any documentation is unusable after all.)

Yet another thing: There are many undocumented features for CW test framework (e.g. JS doc has nothing about before and after, which looks like setup and teardown in OO-style TDD but no one knows exactly how to use it).

[kazk]

The pages also need the full list of currently supported extra packages (e.g. Python has numpy, tensorflow, etc... but it's not listed there;

Some packages were randomly added long time ago and doesn't even work :(
Please don't assume undocumented packages will continue to exist, especially those rarely used or installed as dependencies of others. They're not version locked either, so I've been analyzing the output of pip freeze to ensure compatibility in the new runner :|
Also some features are meant to be used by Qualified.io and not available for Codewars at the moment.

I agree that this project should be documented and organized better. I really wish it was, especially since I'm inheriting this and have been struggling for months to develop the new runner that tries to be a drop-in replacement. I can't do much about this repository right now, but I'll try to help whenever I can by answering questions and posting updates.

[Bubbler-4]

Please don't assume undocumented packages will continue to exist, especially those rarely used or installed as dependencies of others.

Then I believe the relevant packages will be:

• numpy (1.14.0), scipy (1.0.0), and scikit-learn (0.19.1) for various numerical analysis and stats (numpy is used for fast matrix operations too)
• bs4 (4.4.0) (aka BeautifulSoup4) for html parsing and manipulation
• pandas (0.22.0) for data processing? (I don't think I've seen it on CW yet, but it's quite popular in data science, so maybe it's feasible for a kata idea)
• six (1.11.0) for py2-py3 compatibility (This is practically a must for the upcoming Python test framework)

Nothing else will be really necessary (unless Qualified.io says otherwise), though you may check out gmpy2 (2.1.0a1) and sympy (1.1.1) which may be useful for advanced math.

The version numbers are the latest ones right now, and I believe that none of these packages has gone through major API changes, so you may lock the versions as listed here (except possibly SciPy which has bumped its major version this year; the final 0.x should be fine as an alternative).

---

Well, this discussion looks off-topic, but the point is, these info should be in the CW wiki (with the version number and doc links like I did here) if we fix the packages and versions to include in the docker.

[Voileexperiments]

^ Don't forget ada. There are already katas that uses it: https://www.codewars.com/kata/building-an-adaboost-model-with-sklearn-introductory-machine-learning

Also, it's not added yet, but I'd really like to add z3 to current Python environment. Playing with SAT solvers is about as fun as playing with tensorflow.

[kazk]

For Python 2.7 and 3.4, I'm using the versions from pip freeze because currently the submission is considered a failure if anything is written to stderr meaning deprecation warnings are breaking changes for us :|
For 3.6, I'm doing the same except those installed through apt like python3-numpy. Those are installed for Python 3.4 and have issues with 3.6 anyways (#367) so I'm going to install the latest versions. I'll publish the table of installed packages with versions after the migration.

The container image I'm creating at the moment is mainly for backwards compatibility, so I'm not going to add anything extra. But after the migration is complete, I'm thinking of proposing separate images (built on top of the basic language images) tailored to specific areas with additional packages installed.

I'm also working on improving the cost efficiency so that users can get consistent and improved performance while keeping the cost at similar level. This is a required step in order to support advanced challenges that requires more resources.

---

I just approved that sklearn kata, it only requires scikit-learn (numpy, scipy).

[Unihedro]

For the program input, when EOF is reached, note that a value of -1 is stored in the cell under the pointer, which is considered a standard implementation, albeit not the most popular one. Keep that in mind when writing your BF programs as programs that assume otherwise is likely to cause an infinite loop.

From my experience, I don't think this is a reliable nugget of information to rely on. It seems while reaching -1 almost certainly means EOF, there are often unwanted characters before then (I think the \n character?). BF is already hard enough on its own, having users implementing perl's chomp functionality would make it unnecessarily complicated.

It would be nice if the BF environment actually gets changed to feed exactly the given string in the method (so there are no redundant non-print chars) and then 0, which while isn't a standard implementation is often cited for how much sense it makes. Both this online interpreter, this visualizer and this ide writes 0 to the cell when the input is absent or terminated.