Write a proper specification

[robla]

Folks who have been around the electoral reform community for a while seem pretty interested in this format, but as of 2021-06-06, there isn't really a specification. There's only a few wiki pages, some test cases and a few online discussions about the format. It's really similar to ad hoc formats that have been around for 25 years or so, but we should actually have a written record of what we're up to.

[carlschroedl]

I echo @simberaj' call for a BNF (or similar) specification

[simberaj]

I have attempted to write a parser implementation fully compliant with the current ABIF test suite. It can be found at https://github.com/simberaj/votelib/blob/abif/votelib/io/abif.py; the test suite is at https://github.com/simberaj/votelib/blob/abif/tests/io/test_abif.py. The loaders produce Python dictionaries used in the rest of the library to represent votes, mapping preferences to counts. Implementing the spec, I used a couple of assumptions that seem natural to me and IMHO should be included in the spec; I'm listing those below. Wherever I was not so sure, I'm opening a new issue in this tracker.

• Disallow uncounted ballot lines (i.e. each ballot line should start with a vote count number, even if that is 1)
• Whitespace is insignificant anywhere outside square brackets (i.e. full candidate names).
• Scored votes (with slashes) should contain each candidate at most once.
• Outside square brackets, only ASCII alphabetic characters ([A-Za-z]) are allowed as candidate tokens, i.e. no spaces.

Any comments on the implementation (as opposed to the spec itself) more than welcome in votelib's issue tracker at https://github.com/simberaj/votelib/issues/51!

[robla]

I also wrote a parser using Lark. It took me a little bit longer to write the parser than I liked, mostly due to me getting distracted with other matters rather than any problem with Lark. I'm happy with the direction this takes my work in. Lark is a Python library that uses EBNF as its input. Though Lark is specific to Python, the EBNF format is a language-agnostic format which satisfies the request for a BNF specification associated with ABIF.

The first version of the EBNF is available here: https://github.com/electorama/abif/blob/main/abif-v0.01.ebnf

[simberaj]

Hi @robla, I have gone through your EBNF finally. A great job indeed! It seems this could indeed serve as an important part of the specification of ABIF, but I have some questions regarding minor differences between it and my understanding as reflected in the votelib implementation:

• The candidate token definition formats differ: yours seems to accept =A: [Vít Rakušan] while votelib expects [Vít Rakušan]: A, which is IMHO simpler while retaining the benefit of being able to determine the line type from the first byte
• The ballot line seems to accept mixing of commas with other preference separators, like A, B > C = D. Is this desired and if so, what is the intended semantics?
• The JSON spec seems to be a bit restrictive to me (even though we agree the accepted JSON should be a subset of NDJSON, I would at least allow non-string atomic values) while it seems to allow unquoted keys, which is not permitted by JSON; having said that, this is probably the least essential part of the spec for the time being :-)

Some formal remarks:

• I would not use named constants for simple literals (COLON, ASTERISK, etc.) for clarity, but that might be my personal preference.
• I would rename the abifline token to something reflective of the fact it might span multiple lines.

[robla]

@simberaj - Thanks for the reminder to work on ABIF! I've let myself get distracted by other matters (e.g. I started doing some serious Perl development for the first time in years). I think I understand what you're suggesting, but I think I want to follow the following steps before making changes:

1. Create https://github.com/electorama/abif/blob/main/testfiles/test015.abif , which expresses the invalid syntax
2. Add "test015" to https://github.com/electorama/abif/blob/main/abif_test.py , which will (initially) fail
3. Fix my test suite (and the abif.ebnf file) so that running pytest from the top-level directory doesn't return any failures. Of course, there will be invalid ABIF files in the "testcases/" directory, but that's kind of the point of a good test suite.
4. Declare the specification done, because the test suite is done! :rofl:

Seriously, though, I'll take a closer look at things later. I've got a few other things to take care of in my personal life that are interfering with my Python-scripting time, but I'll hopefully have more time very soon. We should also work out how issue #16 is going to work, so that I can start accepting pull requests.