[Native Image] Printing in Clojure on Windows uses wrong charset

[borkdude]

Describe the Issue

When printing in Clojure via the prn or println function, the characters get mangled in Powershell 7 on Windows due to a wrong encoding (Cp1252).

Even when setting: $OutputEncoding = [Console]::OutputEncoding = [System.Text.UTF8Encoding]::new() in powershell and even when calling native-image with "-Djava.file.encoding=UTF-8" "-J-Djava.file.encoding=UTF-8" arguments either at build time or runtime.

Using the latest version of GraalVM can resolve many issues.

• I tried with the latest version of GraalVM.

GraalVM Version

25 LTS

Operating System and Version

Windows (windows-2022 on Github Actions)

Troubleshooting Confirmation

• I tried the suggestions in the troubleshooting guide.

Run Command

./repro
./repro -Djava.file.encoding=UTF-8
$OutputEncoding = [Console]::OutputEncoding = [System.Text.UTF8Encoding]::new()
./repro
./repro -Djava.file.encoding=UTF-8

Expected Behavior

I expect to see "λ⚙️中文" to be printed in all cases. I want to enforce UTF-8 encoding no matter what is the case on the host system.

Actual Behavior

Given this repro program, I see this output:

(ns my.repro
  (:gen-class))

(set! *warn-on-reflection* true)

(defn -main [& _args]
  (prn (System/getProperty "java.file.encoding"))
  (prn (.getEncoding ^java.io.OutputStreamWriter *out*))
  (prn "λ⚙️中文")
  (alter-var-root #'*out* (constantly (java.io.OutputStreamWriter. System/out)))
  (prn "λ⚙️中文"))

"Cp1252"
"?????"
"λ⚙️中文"
nil
"Cp1252"
"?????"
"λ⚙️中文"
nil
"Cp1252"
"?????"
"λ⚙️中文"
nil
"Cp1252"
"?????"
"λ⚙️中文"

Steps to Reproduce

A full reproducer is available in the following repository:

https://github.com/borkdude/graal-repros/tree/init-at-build-time-output-writer-repro

Note the init-at-build-time-output-writer-repro branch.

Just follow the .github/worksflows/main.yml to see the steps to reproduce.

Additional Context

Clojure uses a java.io.OutputStreamWriter on System.out for printing. This writer it set to a var *out*.
When *out* is replaced with a similar writer at runtime, the characters print using the right encoding.
It might have something to do with build time initialization, but I can't reproduce it using a plain class that I init at build time, see Repro.java.

Run-Time Log Output and Error Messages

No response

[jovanstevanovic]

@pejovica, I’ve stumbled upon this issue. Do you have any idea what might be causing it?

[selhagani]

Hi @borkdude,

Thank you for reaching out to us about this issue.

Perhaps it is simillar to the following issue #11214

[borkdude]

It looks related, but I'm not sure if it's the same issue or root cause

[christianhaeubl]

Most likely the same root cause as #11214.

@borkdude : below is a list of relevant system properties that you can try setting to UTF-8 at build-time (via -J-D...):

• file.encoding
• sun.jnu.encoding
• native.encoding
• stdout.encoding
• stderr.encoding
• stdin.encoding

Based on your usage, I would assume that the issue is gone once you set stdout.encoding at build-time.

[borkdude]

@christianhaeubl I accidentally built my binary with "-Dstdout.encoding" which appears to reset this property to an empty string which curiously enough also fixed the problem :-)