Clojurians Log v2

Likewise. Windows 10 laptop with Powershell clj installed. If you provide detailed instructions to attempt to repro, I'll try it...

with a main.clj like:

(ns main)

(defn -main
  [& args]
  (prn args))

This is based on https://github.com/juxt/pack.alpha/issues/40#issuecomment-513471544 where the trace arguments is done like this https://github.com/juxt/pack.alpha/commit/417fc0ff0bc620a0dfaf5ce907436d5987e2fb59

PS C:\Users\seanc\clojure\dom> clj -m example.main -C:dev
(-C: dev)
PS C:\Users\seanc\clojure\dom> clj -m example.main '-C:dev'
(-C:dev) 

I guess I'm not surprised that Windows needs the argument quoted really, given some of the weirdness of invoking PS scripts.

I guess the weird thing for me is that clj works properly here? e.g. for clj -C:dev -m example.bar

out of curiosity, is this isolated to -C:?

I'm still uncertain whether this is a powershell issue or a clj issue. I'm leaning towards the latter due to the fact that clj -C:dev is parsed correctly (I'm assuming at least!)

It's to do with how main opts are passed through the running program.

PS doesn't know what you're doing so it parses the arguments for you -- the ' around it defeat that.

If you try to specify deps on the cmd line for clj on PS, you need to triple up the double quotes around the version as I recall... it's definitely annoying.

Just so I'm certain though, clj -C:dev results in the expected arguments going to the tdeps classpath calculator?

But clj -C :dev is also valid.

For sure, but isn't what is written in most documentation.

I'm still confused as to why PS decides to parse arguments in [main-opt] but not to clj itself.

Right. But that's how it works on windows

It parses them in both cases

clj -C :dev does not work on Mac / Linux but it does work on Windows

If you need -C:dev passed into your code, you need '-C:dev'

it does work on mac/linux 🙂

are you saying that when using clj on windows, you always put the space between, e.g. clj -C :dev?

When I type clj -C :dev on Linux it doesn't work

I get Missing required argument for "-C ALIASES" on Linux

On windows it behaves exactly like clj -C:dev

Another Windows PS weirdness with quoting

PS C:\Users\seanc\clojure\dom> clj -Sdeps '{:deps {seancorfield/next.jdbc {:mvn/version """1.0.2"""}}}'
Downloading: seancorfield/next.jdbc/1.0.2/next.jdbc-1.0.2.pom from <https://repo.clojars.org/>
Downloading: seancorfield/next.jdbc/1.0.2/next.jdbc-1.0.2.jar from <https://repo.clojars.org/>
Clojure 1.10.1
user=&gt; ^Z
PS C:\Users\seanc\clojure\dom&gt; clj -Sdeps '{:deps {seancorfield/next.jdbc {:mvn/version "1.0.2"}}}'
Error while parsing option "--config-data {:deps {seancorfield/next.jdbc {:mvn/version 1.0.2}}}": java.lang.NumberFormatException: Invalid number: 1.0.2
PS C:\Users\seanc\clojure\dom&gt;

Crazy CR/LF fixed!

I think it used to work on Linux. I agree it doesn't now. It's weird because it's just tools.cli in use.

No, not for the shell script itself (`clj`)

ah, there's preparsing. I see.

Just so I am 100% certain I understand. Is there something in clojure.ps1 that converts -A: foo into -A :foo? Or in windows is it just expected that users will type something different?

I didn't say -A: foo is treated like -A :foo.

I think there are several things at play here. clj allows for the leading : to be omitted on an alias.

(on all platforms)

So clj -Cdev and clj -C:dev are treated the same -- by clj.

Windows PS parses -C:dev as -C: dev and then ignores the the : after -C

so clj sees -C dev

Ah!

so to match tdeps behavior in pack I need to also ignore the trailing : somehow. :thinking_face:

That isn't tdeps behavior, per se.

I suspect that is PS script behavior.

if it was general PS script behavior, then it would also happen in the argument passed to pack.

But I thought you said that -C:aot was being passed to pack as -C: aot separately?

And the println example shows that -C:dev is parsed -- by PS -- as -C: dev two separate args

It is being passed to pack separately. Did I misunderstand you, I thought you meant that it was general PS script behavior, do you mean that it's clojure ps script behavior?

I think it's PS behavior.

elseif ($arg.StartsWith('-R')) {
      $aliases, $params = $params
      if ($aliases) {
        $ResolveAliases += ":$aliases"
      }
    }

But it doesn't matter with clj because a) the leading : can be omitted on aliases because that's what clj does on all platforms and b) the trailing : doesn't matter in the PS script for clj.

Right. The clj script only cares about the leading character.

I wonder if this works on windows: -Ranything foo

My point is that Windows users of clj need to get used to "weird" quoting and already have to triple-quote versions in -Sdeps for example.

PS C:\Users\seanc\clojure\dom> clj -Cand_anything_you_want dev
Error building classpath. Specified aliases are undeclared: [:dev]
PS C:\Users\seanc\clojure\dom>

cool, it works

Here's the difference between the -Sdeps case and the -C: case

clj -A:pack -m pack.main -C:foobar
      ^ works on win       ^ does not work on win

fwiw: https://clojure.atlassian.net/browse/TDEPS-133

@dominicm Because that cmd line is actually passed in as

clj -A: pack -m pack.main -C: foobar

It's absolutely consistent.

@seancorfield I know, but then clojure.ps1 massages it to -A:pack -m pack.main -C: foobar So the massaging happens in one place but not the other. Which makes perfect sense now I've read the source.

No, it doesn't.

The PS script knows that the arguments are parsed and passed in separately.

PS C:\Users\seanc\clojure\dom> clj -Cand_anything_you_want:dev
Error building classpath. Specified aliases are undeclared: [:dev]
PS C:\Users\seanc\clojure\dom>

isn't that what this code does?

elseif ($arg.StartsWith('-R')) {
      $aliases, $params = $params
      if ($aliases) {
        $ResolveAliases += ":$aliases"
      }
    }

The cmd line is parsed to clj -Cand_anything_you_want: dev by Windows/PS.

The clj shell script checks the first two chars (`-C`) and ignores the rest.

The shell script gets the next argument as the params for that.

So the issue is that tools.cli does not accept -A: aot as being the same as -A:aot and so it parses it "incorrectly"...

So... as I said... Windows requires extra quoting to override the PS parsing... so that strings are passed into tools.cli (in pack) as-is.

Communicating on the internet is hard 😄 I am confident I understand the behavior now, and I think you do too. I guess my wording isn't communicating what I'm trying to say very well.

(and it's no surprise because you already have to say """1.0.2""" to get "1.0.2" in -Sdeps on Windows)

on ps

My suggestion: tell Windows users of pack to wrap options in ' quotes '

that makes copy/paste harder. I was actually thinking of implementing the ps1 code above at the java-level. By aliasing -R: to -R.

cmd users do not have this problem, though they have other ones

Any cmd user trying to run clj is going to have even worse problems @carkh

We're not talking about lein/`boot` here. Just clj.

there is the unofficial command line, works great

I guess that's not my full reason. My full reason is that -Sdeps is an escaping issue with the shell, and if pack took a -Sdeps flag, it would work the same as clj's does. In this case, clj is "fixing" the options to make -R:foo behave like it does on Linux, I'd like pack to look & feel like clj does.