Welcome! Please see the About page for a little more info on how this works.

0 votes
in Syntax and reader by

Currently keywords and symbols containing more than one slash are disallowed by the spec, but allowed by the readers.
This trivial patch makes them unreadable by the readers too.

Pre:

user=> :foo/bar/baz :foo/bar/baz

Post:

user=> :foo/bar/baz RuntimeException Invalid token: :foo/bar/baz clojure.lang.Util.runtimeException (Util.java:221)

9 Answers

0 votes
by

Comment made by: jafingerhut

Perhaps overlap with CLJ-1527 ?

0 votes
by

Comment made by: zamaterian

Please notice that keywords with more than one slash has a different hashcode across clojure version 1.5 and 1.6

This creates a problem when using a datomic version that works with clojure 1.5 under clojure 1.6 and the schema have one or more keys with more than one slash.

0 votes
by
_Comment made by: zcaudate_

Please reconsider this `fix` for the following reasons:

- (keyword "foo/bar/baz") will also need to be fixed for consistency reasons
- It will break users of my code
   - http://docs.caudate.me/adi/adi-guide.html
   - http://docs.caudate.me/adi/adi-walkthrough.html#schoolyard

Please see discussion on this topic below:


dm3 [5:04 PM]
Is there a reason why `(read-string "a/b/c")` is OK, while `(clojure.tools.reader/read-string "a/b/c")` fails with `Invalid token`?

hiredman [5:04 PM]
there is a ticket to fix read-string

dm3 [5:05 PM]
so the correct behaviour is to fail?

hiredman [5:05 PM]
http://dev.clojure.org/jira/browse/CLJ-1530

dm3 [5:06 PM]
thanks, seems like a breaking change :simple_smile:

hiredman [5:07 PM]
the docs have precluded symbols like a/b/c for some time, and the behavior of how those unspecified symbols were read apparently changed at some point

lucas [5:09 PM]
joined #clojure

zcaudate [5:16 PM]
@hiredman: seriously. I’m really peeved about that because I’ve been using `:a/b/c` keywords for a while… even wrote a whole freaking library to deal with that http://docs.caudate.me/hara/hara-string.html#api---path (edited)

[5:16]
and now they are taking it out

[5:18]
i’m of the opinion that the `::foo/baz` keywords should be taken out first

[5:20]
 ```user=> (require '[clojure.walk :as walk])
nil
user=> ::walk/hello
:clojure.walk/hello
```

[5:20]
that causes way more problems

[5:20]
especially with analysers

[5:22]
https://github.com/jonase/kibit/issues/14
 
GitHub
Kibit breaks on namespace qualified keywords · Issue #14 · jonase/kibit · GitHub
If the code contains namespace qualified keywords with aliases, Kibit errors out with a Invalid token exception. The following code demonstrates the problem - ;;; foo.clj (ns foo) ;;; bar.clj (n...

[5:25]
@dm3 if there’s really a problem, you can patch it:

[5:25]
https://github.com/helpshift/hydrox/blob/master/lein/src/leiningen/hydrox/setup.clj
 
GitHub
helpshift/hydrox
hydrox - Dive deeper into your code

dm3 [5:31 PM]
yeah, would have to patch cljs.tools.reader too unfortunately :confused:

[5:31]
think I'll just work around that

bronsa [5:36 PM]
@zcaudate: `::foo/bar` style keywords are in by design and not going anywhere, `:foo/bar/baz` have always been invalid by the spec and undefined behaviour (edited)

[5:37]
@dm3: is changing undefined behaviour a breaking change? :simple_smile:

dm3 [5:37 PM]
breaking as in breaking people's code :simple_smile:

[5:37]
e.g. zcaudate

bronsa [5:38 PM]
that code is already broken if it's using invalid clojure. It's just accidentaly working

dm3 [5:39 PM]
I'm looking from a pragmatic perspective. Theoretically you're right :simple_smile:

[5:40]
and I'm not judging either

bronsa [5:40 PM]
pragmatically, `:foo/bar/baz` is a bug waiting to happen. what does `(namespace :foo/bar/baz)` return?

dm3 [5:40 PM]
whatever it returns currently?

[5:41]
I mean it's kind of defined by the implementation

sveri [5:41 PM]
@dm3 @bronsa I wouldnt even agree that you are theoretically right. As soon as enough people adapted the broken code it falls under something like a common law that was accepted by both parties for a long enough time.

bronsa [5:42 PM]
what about `namespace` on `(keyword "foo/bar" "baz")` and `(keyword "foo" "bar/baz")`

dm3 [5:43 PM]
I agree that currently implemented semantics are messy

[5:43]
but my point was that it's still a breaking change

[5:43]
not that it's a "bad" change

[5:43]
that's a judgement

bronsa [5:43 PM]
@sveri:  I would agree with you, as long as the undefined behaviour we accept as defined doesn't cause impossible to fix semantics.

[5:44]
that's why for example, the patch that made symbols starting with numbers illegal was rolled back

[5:45]
it broke existing code, it didn't cause weird semantics so it was rolled back. not the case with `:foo/bar/baz`

[5:45]
@dm3: you could make the point that fixing any bug is a breaking change then -- people might be relying on that bug.

dm3 [5:46 PM]
yep, I guess what matters is how obvious the incorrect behaviour is and how many people rely on it

bronsa [5:46 PM]
if the doc explicitely says "you can use ​*one*​ `/` inside a symbol", then if you're using more than one you're writing invalid clojure and you should expect it to maybe break (edited)

dm3 [5:47 PM]
I really haven't even thought about multiple slashes (nor noticed the docs) in a symbol in ~3 years of using Clojure

[5:47]
my initial thought (today) was that it was permitted

[5:47]
and the namespace would be the first segment before the first slash (edited)

bronsa [5:51 PM]
well, that doesn't make much sense though. `/` in clojure means `namespace separator`. if I see `FOO/BAR`, no matter what `FOO` and `BAR` are, I know that `FOO` is the namespace, and `BAR` is the name. if you want to express paths with keywords as in @zcaudate's lib, you should use a different separator in your keywords that doesn't have a special meaning in clojure, like `.` (i.e. `:foo/bar/baz` -> `:foo.bar.baz` or `:foo/bar.baz`) (edited)

dm3 [5:51 PM]
I do not want to argue semantics. Just sharing one point of view

bronsa [5:52 PM]
and my point is that pragmatic point of views (especially when they go against the current doc) should only be considered if the semantics they imply are clear and unambiguous

sveri [5:53 PM]
@bronsa: Nice explanation, thank you :simple_smile:

dm3 [5:53 PM]
yep :simple_smile:

[5:54]
I agree to that, as you have to make decisions in the end


zcaudate [9:07 PM]
@bronsa: written form of communication has a way of making things more serious than they seem

[9:08]
honestly… i knew it was coming since 1.6 when the edn reader started breaking my code

[9:09]
it’s probably more my fault for not communicating this earlier but oh well.. we all have to roll with the times

bronsa [9:09 PM]
@zcaudate: yeah no worries, I was just using your lib as an example since you brought that up

zcaudate [9:10 PM]
having said that, you can imagine my disappointment because I had designed an entire query semantic based on the keyword `:foo/bar/baz` feature (now bug) (edited)

[9:11]
http://docs.caudate.me/adi/adi-walkthrough.html#querying

[9:11]
you noticed I didn’t use `(adi/select ds {:student/classes/teacher/name "Mr. Blair"}})`

[9:12]
in my docs because the tests started breaking

jstew [9:12 PM]
@zcaudate: You put out so much quality stuff that I wonder if you ever sleep!
1  

bronsa [9:12 PM]
@zcaudate: luckly the fix should be easy :simple_smile: just replace `/` with `.`

zcaudate [9:12 PM]
no!

[9:13]
see the problem is… datomic has things like `account.type/user`

[9:13]
and so I would have to do the cljs thing `account.type$user`

bronsa [9:13 PM]
(that's some high quality documentation btw, good job)

zcaudate [9:14 PM]
@bronsa: hahaha thanks… so maybe you can push the fix to 1.10

[9:14]
that way I can get a few more months left

[9:15]
like it’s not a big deal… but I thought that there is a parallel between the path structure of the `/` calls and the nesting of maps

[9:16]
and so there is an equivalence to `{:student {:classes {:teacher {:name '(?fulltext "Blair")}`

[9:17]
and `{:student/classes/teacher/name "Mr. Blair"}`

[9:17]
which is prettier in my opinion

bronsa [9:17 PM]
@zcaudate: sorry if that wasn't clear, but I don't actually have any control over when or what gets into clojure or not, I'm just a contributor :simple_smile: so there are chances that the clojure/core team will take a different decision and actually decline that ticket (I would be really disappointed if that was the case though!).  If that will happenI will obviously make a change to `tools.reader` to allow them aswell, (edited)

zcaudate [9:18 PM]
@bronsa: damn.

[9:19]
well… maybe you can highlight this fact

[9:19]
and also if the fix is made, a fix to `(keyword "foo/bar/baz")` will also be needed (edited)

bronsa [9:20 PM]
don't think that'll ever be done. validating inputs to `keyword`/`symbol` etc has been asked/discussed tons of times and repeatedly declined for performance reasons

zcaudate [9:20 PM]
so that’s a matter of consistency.

bronsa [9:20 PM]
(not that I agree with that decision, but it seems like Rich isn't going ot change his mind on that)

[9:21]
@zcaudate: there's a difference between what a symbol/keyword can be at runtime, and what a valid read-time symbol/keyword is

zcaudate [9:21 PM]
and also, it means I can setup a reader macro #k foo/bar/baz and get the same effect

[9:22]
like it’s stupidly ugly but i believe it will work

bronsa [9:23 PM]
but the ambiguity of what to do with `namespace` and `name` still remains so dunno

[9:23]
@zcaudate: that wouldn't work either way, if http://dev.clojure.org/jira/browse/CLJ-1530 gets accepted neither `:foo/bar/baz` ​*nor*​ `foo/bar/baz` will be valid anymore

[9:24]
@zcaudate: btw I'd suggest you log your issues with that ticket in a comment there if you feel strongly against it

new messages
[9:25]
I ​*suspect*​ that the response will be "you should use a delimiter that doesn't have a special meaning in clojure", but I might be completely wrong (I find the core team doesn't agree with my opinions quite frequently :) ) especially if you point out that your library will break. (edited)

zcaudate [9:30 PM]
@bronsa: thanks for the heads up. I’ll leave a comment and add a prayer for the bdfl
0 votes
by

Comment made by: alexmiller

Chris -

  • (keyword "foo/bar/baz") will still be fine. Programmatic keywords can be created for any string - this is intentionally much broader than what the reader supports as a literal in code and it's a feature that's widely used. At some future point, there may be an escaping mechanism for symbols or keywords with characters outside the spec such that the reader could read them as well, but that's outside the scope of this.
  • Your api is using illegal keywords according to http://clojure.org/reference/reader and you should not expect them to work. I think you should change your library.
0 votes
by

Comment made by: zcaudate

Alex

  • I wouldn't necessarily call it illegal as the current behavior in the edn.reader was added in 1.6 without warning.
  • Also, if (keyword "foo/bar/baz") is allowed to exist, then there still would be the indeterminate namespace/name problem that @bronsa highlighted. I would argue for consistency and if :foo/bar/baz is illegal in the reader, then it should be illegal everywhere
  • My library should be fine... but users of the library may have to change their queries
0 votes
by

Comment made by: alexmiller

The reader page clearly states "'/' has special meaning, it can be used once in the middle of a symbol to separate the namespace from the name" and keywords are "like symbols". This has been on the reader page since the oldest version I can find in the internet archive (July 2008). edn (despite its similarities) is a separate thing than Clojure, and irrelevant.

The 2-arity form of {{keyword}} can be used without ambiguity: {{(keyword nil "foo/bar/baz")}}. The 1-arity form will split based on the first / found (in this example into "foo" and "bar/baz"). I see no reason that would need to change.

0 votes
by

Comment made by: zcaudate

I disagree on the "specialness" of / in keywords, especially if it is allowed as a string.

This leads to another problem in the ambiguity of output:

user=> (keyword "foo" "bar/baz")
:foo/bar/baz

user=> (keyword nil "foo/bar/baz")
:foo/bar/baz

Unless the output is displayed as :foo//bar/baz, it is unclear where the namespace is and the concept of code is data will be diminished if the output cannot be read back as data.

There is also the case for symbol:
user=> (symbol "foo/bar/baz")
foo/bar/baz

well... at least all the examples listed are all consistently "illegal"

Anyways, even if the docs had not explicitly stated such, :foo/bar/baz has existed since the beginning of clojure and personally, it seems to be more string-like than symbol-like. Ironically, I'm pretty sure that I got the idea of using multiple slashes in keywords from reading the datomic documentation from back when I started work on adi.

Ultimately, the decision is not mine to make and I do value the guidance of clojure team over the development of the language. I do however, hope that my points for keeping things as they are can be recognised and be taken into consideration.

0 votes
by

Comment made by: zcaudate

I'll quote Rich https://www.youtube.com/watch?v=P76Vbsk_3J0 @ 5:10

"Many of the things you consider to be problems (with lisp) are features... down the line..."

0 votes
by
Reference: https://clojure.atlassian.net/browse/CLJ-1530 (reported by bronsa)
...