Clojurians Log v2

proposed alternative name:

:alias-same-as-ns

That makes me think it may lint for this:

(ns foo (:require [bar :as foo])) ...

:alias-same-as-aliased-ns

to be clear, I think :alias-same-as-ns could be misinterpreted as meaning :alias-same-as-this-ns

I'm not worried about that, an alias is for an ns, I think it's clear enough

OK. I’ll give it a few days to make sure this is settled before doing the renaming 🙂

probably releasing clj-kondo on Monday

this is the last bit that I'll merge

The message pretty much says the same thing: > alias same as namespace name Btw, the style for most linters are:

Alias same as namespace name: 

alternative wording: :alias-equals-ns-name or

:alias-equals-ns

for whatever it is worth: I just implemented my own linter to yell at us if we use :as on short namespaces (2 segments or less disallow aliases). Doing so was partially motivated by [foo :as foo] being found more and more in the codebase as llms churn out code and folks miss things like this while reviewing

2 segments or less disallow aliases? what is a segment? can you give an example? foo.bar.baz is three segments?

That's the usual parlance isn't it?

I guess so, I was just surprised that people use segments in aliases at all that much

I think what they mean is that they are forbidding [foo.bar :as bar]

oh I re-read and now I see it

so (clojure.test/is ...) instead of (t/is ....)? that seems... tiring :)

Wow, I alias 2-segment nses all the time... and a 1-segment ns is problematic for other reasons.

tbh I like last segment aliases the best

maybe we could get back to the original issue since I don't like it but I don't want my personal taste guide every possible thing in clj-kondo

Makes me curious how often folks use 2+ segment aliases tho'? I really don't like those but I have seen cases where they are better for readability than inventing multiple similar aliases...

Perhaps a more realistic example for discussion would be (:require [my.ns :as my.ns]) then?

Yes, things like that would be what this linter would guard against to my understanding.

Yeah, I'd be in favor of a linter that flagged :as having the same alias as the namespace in that case.

(:require [my.ns]) generally indicates an ns loaded for side-effects (and therefore suppresses the "unused ns" linter). I think the "breaks the visual pattern" aspect there is important, since the unaliased ns is "special".

that's a very good point!

it took a little while to get used to this and I also originally somewhat rebelled against it. then you get used to it and start thinking it ends up being nice. So then you make robots enforce it so tooling sticks with your conventions 🙂

It's fascinating sometimes the weird little quirks individual teams fall into over time with Clojure... 🤔

not at all suggesting this becomes a built-in. Just a data point towards flagging foo :as foo

indeed. I've been writing clojure for a long time and first time I've come across this convention

@seancorfield you had an argument against this linter, yet you gave it a thumbs up. wanna explain?

I don't think I was arguing against the linter per se -- I was arguing against single-segment nses 🙂 and I was arguing in favor of the [my.ns] require form as specifically not needing :as my.ns because I think the argument that it "breaks the visual pattern" is wrong.

thanks for clarifying your position

What wasn't clear to me: you recognize that [my.ns] is usually used for side effects. But would you still be in favor of [my.ns] if it's not for side effects, but people actually use (my.ns/foobar)? I can see one downside of this: when removing (my.ns/foobar) it won't be clear thereafter, if the namespace was loaded for side effects or not, or simply unused.

> you recognize that [my.ns] is usually used for side effects Correct. > you still be in favor of [my.ns] if it's not for side effects I wouldn't do it, no. I'd add :as ns (assuming "ns" is some real ns name, not literally ns) I really, really don't like seeing (my.ns/foobar) in code.

And I really don't like dotted aliases.

then you are not in favor of the original issue, I think

If pushed, I might do (:require [my.ns :as my-ns])

at least, not in general. just when the namespace was used for side effects, but we already have that convention in clj-kondo, so that's not what the issue poster had in mind

The :redundant-alias would flag [my.ns :as my.ns] (good: I don't like that) but it would not flag [my.ns] (good: use for side-effects) or [my.ns :as my-ns] (good: non-redundant alias that is also non-dotted).

exactly. and this is why it's even encouraged to use an alias for a namespace that you're going to be calling functions from, rather than using the ns name directly. the proposed issue breaks that pattern

so this is why I was confused about your 👍

:redundant-alias suggests that you would remove the alias, which is clearly not what we want for this pattern. redundant = remove in clj-kondo.

I think using (my.ns/foobar) after (:require [my.ns]) is a separate issue and would warrant a separate linter at some point maybe.

This is one of those tough spots where there's a real subtlety in usage and intent. In my code, I'd never trigger :redundant-alias but I'd be happy for it to be flagged in, say, AI-generated code.

For example, I'd also welcome a :no-dotted-alias linter since I do not like that style.

would you like the AI to switch to (:require [foo.bar]) + (foo.bar/baz) after seeing the warnings? because I think that's what's going to happen if the linter is called :redundant-alias :)

I'd want (:require [foo.bar :as bar]) or (:require [foo.bar :as foo-bar])

right. so you are in favor of the linter, but the name should probably change to :alias-same-ns-name or whatever (can't come up with a good one right now).

as in, the hint on how to deal with the issue, should be more neutral

💤 now

a nice vaguely related issue I once solved: https://github.com/clj-kondo/clj-kondo/issues/864