Clojurians Log v2

clojure -A:foo and clojure -M:foo without main-opts do the same thing: invoke a REPL with alias foo merged in -- right? Should -M:foo give a warning when you forget to supply main-opts in deps.edn maybe?

or should -A not invoke a REPL since that's a clojure main invocation?

-A is for REPL invocation, according to the docs.

so -A and -M overlap there I guess

I suspect that's a bit of a grey area given that :main-opts ["-r"] works to run a REPL, yes?

well, I invoked an alias with -M while I forgot to provide :main-opts in deps.edn, so it invoked a REPL for me

that's what prompted these maybe too far fetched thoughts

Yeah, I would sort of expect -M to at least give a warning if no :main-opts are provided. @alexmiller?

(! 651)-> fgrep :repl deps.edn 
  :repl {:main-opts ["-r"]}
(! 652)-> clj -M:repl
Clojure 1.10.1
user=> ^D
(! 653)-> clj -A:repl
WARNING: Use of :main-opts with -A is deprecated. Use -M instead.
Clojure 1.10.1
user=> ^D

(so, yeah, you can specify -r explicitly as a :main-opts item)

And the corollary:

(! 656)-> fgrep :dummy deps.edn 
  :dummy {}
(! 657)-> clj -M:dummy
Clojure 1.10.1
user=> ^D
(! 658)-> clj -A:dummy
Clojure 1.10.1
user=> ^D

you can supply main-opts on the command line, so -M does necessarily have to have :main-opts

clj -M:my-alias-with-just-deps -m my.ns

Yes, but that's not the case we're discussing here.

well, what is?

See the two console examples I pasted. No command-line main-opts there.

-M invokes clojure.main. clojure.main does not require opts.

Fair enough. The "implicit" -r option 🙂

Why draw the distinction in the docs then that -A is for REPL execution?

because -M might not invoke a REPL when main-opts are given

really, it can be used to supply aliases for anything but it's only useful with REPL (because -M can be used for main and -X can be used for functions)

(and for allowing legacy stuff to continue working)

I guess this will become less confusing once -A stops executing :main-opts

yes

I'm curious how you plan to distinguish between -A for a REPL and -M executing clojure.main which might pick up other main opts?

I would maybe prefer -M to trigger a warning when there is no :main-opts [] when you want to explicitly pass 0 arguments to main (i.e. invoke a REPL)

@borkdude I don't think that is going in the right direction...

what do I need to distinguish?

It seems that the only way to avoid continued confusion is to have clojure.main without arguments not start a REPL -- and require -r for that? 🙂

I don't follow

I mean: how is -A going to start a REPL without running clojure.main?

never mind

it already does this

it invokes clojure.main

that's an implementation detail

clj without -M or -X will start a repl, you let me figure out how to do that :)

yes?

I guess you could always have a new entry point for a REPL and have -A invoke that...?

I don't understand what problem is bothering you

(or rather "have the absence of -M and -X invoke that")

the absence of -M or -X means: invoke a repl

in other words, the default behavior of clj is: invoke a repl

(! 659)-> clj -A:dummy -r
WARNING: When invoking clojure.main, use -M
Clojure 1.10.1
user=> ^D

ok, you're outside the bounds of documented behavior

that's legacy support

-r is a documented main-opt that causes clojure.main to start a REPL, isn't it?

yes, but when we stop accepting main opts outside of -M, that's no longer a thing

there may be some other set of accepted arguments in the repl case that have nothing to do with clojure.main

OK, so there will need to be a change in the behavior of clojure.main at some point to "fix" this?

no

clojure.main's help says

-r, --repl          Run a repl

"clj starting a repl" is distinct from clojure.main

yes, the mechanism clj currently uses to start a repl is clojure.main

but that's an implementation detail

So there will be a non-`clojure.main` entry point that clj can use to start a REPL?

(otherwise this confusion is going to continue)

what I'm saying is - I am promising clj will start a repl. the end.

and I'm promising that clj -M will run clojure.main

and that clj -X will invoke a function with a map

but you're not promising that clojure.main will not support -r at some future point :rolling_on_the_floor_laughing:

there are no plans to make clojure.main do anything different than what it does now

in the before times, clj always ran clojure.main. that is no longer the model

the new model is: • by default, run a repl • with -M, run clojure.main • with -X, execute a function with a map

and "run a repl" does not say anything about clojure.main

if you want to explicitly run clojure.main to run a repl, then clj -M or clj -M -r will do that

The confusion arises because the docs talk about "using -A (for REPL invocation)" when it still runs :main-opts!

well, we're in a bridge period between the two models

The sooner -A stops running :main-opts the better...

(mind you, even in that future world, I think @borkdude’s confusion will still be present, yes?)

if the confusion is that you can use clojure.main to run a repl, then yes

but it's easy to reason about if we are properly conveying that -M == run clojure.main

Argh, I'm being bitten by caching here 😞

(! 665)-> clj -Sforce -A
Missing required argument for "-A ALIASES"
(! 666)-> clj -Sforce -A -r
Missing required argument for "-A ALIASES"

^ That makes more sense.

-M and -X can be invoked without an alias, -A cannot.

what does that have to do with caching? -A always requires an alias

Because this happened before I added -Sforce:

(! 662)-> clj -A
Clojure 1.10.1
user=> 
(! 663)-> clj -A -r
WARNING: When invoking clojure.main, use -M
Clojure 1.10.1
user=> ^D

(note the command history numbers there!)

oh, well that's just a bug then

Here's the repro for that bug:

(! 674)-> clj -Sforce -A
Missing required argument for "-A ALIASES"
(! 675)-> clj -A
Clojure 1.10.1
user=> 

fixed

So the point earlier was: invoking -M with an alias that doesn't have :main-opts assumes that :main-opts is implicitly []. My question was: should this be less implicit, to avoid people ending up in a REPL when invoking -M with an alias that wasn't intended for -M at all

no, because :main-opts is not required

ok

@borkdude in the following :test has no :main-opts -- which of these would you expect to get warnings on and why?

(! 676)-> clj -Sforce -M:test -r
Clojure 1.10.1
user=> ^D
(! 677)-> clj -Sforce -A:test -r
WARNING: When invoking clojure.main, use -M
Clojure 1.10.1
user=> ^D
(! 678)-> clj -Sforce -M:test 
Clojure 1.10.1
user=> ^D
(! 679)-> clj -Sforce -A:test 
Clojure 1.10.1
user=> ^D

we could do something like - if you supplied -M but did not supply either :main-opts or main-opts on the command line, then either warn or error at that point

but honestly that's difficult to check right now due to where that knowledge of different parts of that is

yes, that's exactly the case I was referring to

So -M wouldn't silently start a REPL unless you said -M -r ? I think that would be reasonable.

(if it could be checked)

and then -M means run clojure.main except for this one case where it fails

yeah, I guess it's not worth changing

(I'd be against an error there but I think a warning to remind folks might be worthwhile)

main opts can be supplied on the command line, so not providing arguments means the same as :main-opts [], but you can also do that with -A which is the preferred way of invoking a REPL maybe

@seancorfield 1.10.1.735 if you want to check the -A thing

But I suspect there's no point trying to second-guess what folks will find confusing until after we have a CLI version where -A no longer uses :main-opts

@alexmiller Thanks for the swift fix!

yes

(! 680)-> clj -Sforce -A
-A requires an alias
(! 681)-> clj -A
-A requires an alias

I did not take the time to try on Windows, but it might already catch this case due to the different parsing. I'll check that before we get to a stable release

I'm shutting down, have a good Halloween weekend y'all