The complaints about ISO Prolog, and about ISO standards generally not being available publicly / at no cost, resonate a lot with me. If you compare eg. the POSIX and the ISO C development workflows, and their respective end results, there's a world of difference.
"ISO" (International Standards Organization), where you have "national member bodies", should absolutely be a thing of the past for programming languages (or for anything related to computing). My country's national body's own homepage has a huge tirade, aiming to "dispel myths", such as the "myth" that "standards should be available free of charge". Meanwhile, lots of std orgs have published computing standards with various degrees of openness already.
ISO is a relic when it comes to computing. (So are other standards bodies that choose to remain proprietary; like those behind PCI, SCSI, ... Computing is ubiquitous and these bodies should be forced open by law, for the public interest.)
pjmlp 5 hours ago [-]
POSIX is hardly much more open than ISO, it also has its issues, and contributing to OpenGroup is hardly like doing a pull request.
So is Khronos also a relic? After all if the standards are made available, the contributions are "you have to be this tall to play" kind of entry.
I do agree that ISO should improve itself to modern times, though.
TheAceOfHearts 13 hours ago [-]
Maybe someday someone will find the courage to do a full dump of every ISO standard into LibGen or SciHub.
fanf2 18 hours ago [-]
Yeah
For the horrible tedious details, see the “Policy for the distribution of ISO publications and the protection of ISO’s copyright” aka ISO POCOSA 2012.
doesn’t have to be a legal enforcement if vendors just don’t work with bodies that insist on this counterproductive notion of standards as a profit center.
unfortunately i think there us some degree of collusion here. it’s easier to get your existing proprietary standards ratified if there are fewer players in the room and the palms that need to be greased are clearly marked
OneDeuxTriSeiGo 16 hours ago [-]
It's not exactly clear cut. Standards are unfortunately in general quite expensive to produce and maintain.
Software oriented standards are certainly cheaper than metallurgy, machining, manufacturing, building construction, environmental, health & safety, and the other big classes of standards however they still have quite a cost.
Historically the ISO standards development process for software standards (like the C or C++ standards) happened only in small part asynchronously and historically required large, extended-duration committee meetings where all the details were hashed out in person. This process only really started to change during COVID but even then it's still a very in-person synch-heavy process and that's not exactly cheap to run.
And with most standards, the FDIS (final draft international standard) revisions are made public. They can be found online even if they can be annoying to dig up. For 99% of cases the FDIS revision is more than sufficient and is identical to the published standard minus a typo or grammar mistake here and there.
As the average SW dev or engineer of course you don't need to fork over the cost for the published standard but any large company will probably purchase a catalog of standards rather than deal with the overhead of dealing with FDIS (and any legal risk from not following the "true" standard).
It is also worth noting that pretty much every university library (and many public non-university libraries) has some contract or service that provides access to copies of the standard to members free of cost.
convolvatron 2 hours ago [-]
I used to work in standards. my organization supported my salary, my travel, and paid membership fees. yes, the standards body did rent the hotel ballroom, and run a website. but otherwise the task of making the sausage was all volunteer.
PJBaker 17 hours ago [-]
> every single language document uses its own notation, which is more often than not, a dialect of the (Extended) Backus-Naur Form
I came across this recently while writing an article that references Lua, Go and Python (3.8) syntax. Each of them uses a slightly different form of EBNF.
To make them more easily comparable, I wanted to convert all three to the same format. Looking for something fairly standard (not entirely ad-hoc but also not as formal as ISO/IEC EBNF or RFC 5234 ABNF), I came across Wirth Syntax Notation [0] [1]:
syntax = { production } .
production = identifier "=" expression "." .
expression = term { "|" term } .
term = factor { factor } .
factor = identifier | literal | "(" expression ")" | "[" expression "]" | "{" expression "}" .
literal = "\"" character {character} "\"" .
It turns out that the Go specification already uses WSN [2]. I converted Lua and Python, and then could work with all three language grammars in a consistent, machine-readable notation.
IMHO ABNF is just as annoying to read, especially its insistence on '/' instead of '|' for alternation (when | has universally become the "OR" symbol in other languages) and "%x" for hex prefixes. My preference for syntax descriptions is [ ] for "optional", ",,," or "+" for "1 or more", "*" for "0 or more", and "{min,max}" or "{n}" for repetition ranges, which closely mirrors regex syntax; I'm not sure if this has been standardised or even has a name, but I've seen it far more often than ABNF/EBNF.
Pet_Ant 4 hours ago [-]
First thing I noticed was
> rule = definition; comment CR LF
...so it uses CR/LF as a line terminator as opposed to just LF like POSIX does so you are gonna have text editors and other tools just breaking the format on save.
dwheeler 20 hours ago [-]
I'm the author. Ask me anything!
arh68 17 hours ago [-]
Great article. Informative, qualified. I had not realized how splintered the [E]BNF syntax is, like in the way I already knew timestamps are (3339 vs 8601 vs mm/dd/yyyy &c &c).
Q: what's your ideal way to write Unicode characters clearly? In the W3C/XML spec they'll have stuff like [#x200C-#x200D] but I have no idea what those are, without like a dictionary on hand. Points for specificity, but it doesn't scream "readable".
Your point about standards-not-publicly-available is unfortunately similar to, well, laws. In some areas, "the laws" themselves are not public (!) though perhaps it's a digression better to not get into
pedantically, s/unabiguously/unambiguously/g
dwheeler 8 hours ago [-]
In many cases I think the character itself is clearest. Thankfully most tools can handle Unicode today. That's not always unambiguous, so sometimes an annotation may be helpful, and if the spec is freely available you can copy amd paste from it.
ucarion 20 hours ago [-]
Is the main objection to RFC5234 its weird syntax for repetition (*THING, [THING], 5THING)?
dwheeler 8 hours ago [-]
More or less. It's a perfectly workable approach, but it's the reverse of what practically everyone else does, as well as the reverse of regex notation. Common notations should be common.
90s_dev 17 hours ago [-]
What's your favorite programming language and why is it C?
dwheeler 8 hours ago [-]
I like lots of programming languages, I don't really have 1 favorite. I like the relative simplicity of C, and I have long familiarity with it, but its lack of memory safety and endless undefined behavior make it more difficult to safely use than necessary.
teddyh 20 hours ago [-]
Is there any documented version of EBNF suitable for general use?
Commentaries say its different, but the differences might not matter?
alfiedotwtf 17 hours ago [-]
So if not EBNF for defining grammars, what should people be using ie what would be your preferred that would be powerfully enough and general purpose?
I’ve never seen PCRE used for grammars and not sure if it would be powerfully enough for all cases, but personally I think it would be pretty cool since I find it easier to read than EBNF and it’s so widely used people can slot a specification right into code and it should just work
dwheeler 7 hours ago [-]
One of those alternative specifications is in the W3C Extensible Markup Language (XML) 1.0 (Fifth Edition). As I note, "The W3C specification is much more similar to typical regex syntax making it much easier for today's software developers to understand), avoids the key problems of 14977:1996, and is already clearly described." It's freely available from w3c. You don't need to use XML to use it.
HarHarVeryFunny 20 hours ago [-]
> One of the most common operations in a grammar, by far, is concatenation (aka sequencing). ISO/IEC 14977:1996 requires that a comma be used for every concatenation, so any sequence of N symbols will have N-1 commas.
It seems they were trying really hard to make EBNF NOT look like the grammar it is representing.
> ISO/IEC 14977:1996 represents “one or more” as { symbol }- which means “0 or more symbols, and then subtract the empty set
It's almost as if they were looking for a special syntax to express "1 or more".
Bizarre ... just bizarre.
kstenerud 17 hours ago [-]
These are some of the many reasons why I developed the Dogma metalanguage.
That, and also I needed a language that could describe binary data.
mdaniel 16 hours ago [-]
Do I correctly understand that you don't currently have a parser generator for .dogma files? I tried clicking around in a few of the adjacent repos without much luck
Anyway, my go-to tire kicking for any such binary file description format is parsing .pdf files, since they are ferociously hard, and include backreferences
kstenerud 8 hours ago [-]
I haven't gotten around to writing a parser generator yet. I only wanted this for documentation purposes, and haven't had the spare time to take it further yet.
There are definitely going to be horribly convoluted formats that it can't describe without help (which is where custom functions come in). But it's been able to describe most formats I've thrown at it, so that's good enough for Dogma v1.
I was able to get it to describe minidump without function help...
mannyv 20 hours ago [-]
(4) -> I would say that the lack of regex use is a plus. Yes I use them all the time. For anything complicated I'll just ask ChatGPT to make it. And of course, are regexs all the same across everything?
(5) -> Once you get it you get it.
Every language specification is odd in its own special way. In any case I haven't really seen BNF/EBNF used in the last few years, so this is probably a moot discussion.
alfiedotwtf 17 hours ago [-]
How are the grammars you see on a regular basis specified if not by BNF?
mdaniel 16 hours ago [-]
I'm not the person you asked but I believe there are two broad classes of answers: hand rolled parsers and PEG grammars
"ISO" (International Standards Organization), where you have "national member bodies", should absolutely be a thing of the past for programming languages (or for anything related to computing). My country's national body's own homepage has a huge tirade, aiming to "dispel myths", such as the "myth" that "standards should be available free of charge". Meanwhile, lots of std orgs have published computing standards with various degrees of openness already.
ISO is a relic when it comes to computing. (So are other standards bodies that choose to remain proprietary; like those behind PCI, SCSI, ... Computing is ubiquitous and these bodies should be forced open by law, for the public interest.)
So is Khronos also a relic? After all if the standards are made available, the contributions are "you have to be this tall to play" kind of entry.
I do agree that ISO should improve itself to modern times, though.
For the horrible tedious details, see the “Policy for the distribution of ISO publications and the protection of ISO’s copyright” aka ISO POCOSA 2012.
https://archive.org/details/iso-pocosa
unfortunately i think there us some degree of collusion here. it’s easier to get your existing proprietary standards ratified if there are fewer players in the room and the palms that need to be greased are clearly marked
Software oriented standards are certainly cheaper than metallurgy, machining, manufacturing, building construction, environmental, health & safety, and the other big classes of standards however they still have quite a cost.
Historically the ISO standards development process for software standards (like the C or C++ standards) happened only in small part asynchronously and historically required large, extended-duration committee meetings where all the details were hashed out in person. This process only really started to change during COVID but even then it's still a very in-person synch-heavy process and that's not exactly cheap to run.
And with most standards, the FDIS (final draft international standard) revisions are made public. They can be found online even if they can be annoying to dig up. For 99% of cases the FDIS revision is more than sufficient and is identical to the published standard minus a typo or grammar mistake here and there.
As the average SW dev or engineer of course you don't need to fork over the cost for the published standard but any large company will probably purchase a catalog of standards rather than deal with the overhead of dealing with FDIS (and any legal risk from not following the "true" standard).
It is also worth noting that pretty much every university library (and many public non-university libraries) has some contract or service that provides access to copies of the standard to members free of cost.
I came across this recently while writing an article that references Lua, Go and Python (3.8) syntax. Each of them uses a slightly different form of EBNF.
To make them more easily comparable, I wanted to convert all three to the same format. Looking for something fairly standard (not entirely ad-hoc but also not as formal as ISO/IEC EBNF or RFC 5234 ABNF), I came across Wirth Syntax Notation [0] [1]:
It turns out that the Go specification already uses WSN [2]. I converted Lua and Python, and then could work with all three language grammars in a consistent, machine-readable notation.[0] https://en.wikipedia.org/wiki/Wirth_syntax_notation
[1] https://dl.acm.org/doi/10.1145/359863.359883
[2] https://go.dev/ref/spec#Notation
https://github.com/egberts/vim-syntax-ebnf
And a EBNF format detector:
https://github.com/egberts/filetype-ebnf-grammars
And a master list of all variants of EBNF:
http://www.cs.man.ac.uk/~pjj/bnf/ebnf.html
Here is a list of all variants of EBNFs as well as Vim syntax highlighter for EBNF variants:
https://github.com/egberts/vim-syntax-ebnf
And a EBNF format detector:
https://github.com/egberts/filetype-ebnf-grammars
And a master list of all variants of EBNF:
http://www.cs.man.ac.uk/~pjj/bnf/ebnf.html
> rule = definition; comment CR LF
...so it uses CR/LF as a line terminator as opposed to just LF like POSIX does so you are gonna have text editors and other tools just breaking the format on save.
Q: what's your ideal way to write Unicode characters clearly? In the W3C/XML spec they'll have stuff like [#x200C-#x200D] but I have no idea what those are, without like a dictionary on hand. Points for specificity, but it doesn't scream "readable".
Your point about standards-not-publicly-available is unfortunately similar to, well, laws. In some areas, "the laws" themselves are not public (!) though perhaps it's a digression better to not get into
pedantically, s/unabiguously/unambiguously/g
Commentaries say its different, but the differences might not matter?
I’ve never seen PCRE used for grammars and not sure if it would be powerfully enough for all cases, but personally I think it would be pretty cool since I find it easier to read than EBNF and it’s so widely used people can slot a specification right into code and it should just work
It seems they were trying really hard to make EBNF NOT look like the grammar it is representing.
> ISO/IEC 14977:1996 represents “one or more” as { symbol }- which means “0 or more symbols, and then subtract the empty set
It's almost as if they were looking for a special syntax to express "1 or more".
Bizarre ... just bizarre.
https://dogma-lang.org/
https://github.com/kstenerud/dogma/blob/master/v1/dogma_v1.0...
That, and also I needed a language that could describe binary data.
Anyway, my go-to tire kicking for any such binary file description format is parsing .pdf files, since they are ferociously hard, and include backreferences
There are definitely going to be horribly convoluted formats that it can't describe without help (which is where custom functions come in). But it's been able to describe most formats I've thrown at it, so that's good enough for Dogma v1.
I was able to get it to describe minidump without function help...
(5) -> Once you get it you get it.
Every language specification is odd in its own special way. In any case I haven't really seen BNF/EBNF used in the last few years, so this is probably a moot discussion.
e.g. https://github.com/llvm/llvm-project/blob/llvmorg-20.1.5/cla... and https://github.com/python/cpython/blob/v3.13.3/Grammar/pytho...