From b5256d4bd5aa24f8bad57d18b09ed8d104be351e Mon Sep 17 00:00:00 2001
From: Knut Wannheden <knut@moderne.io>
Date: Thu, 10 Sep 2026 09:06:50 +0200
Subject: [PATCH 1/2] build: pin file.encoding so the Maven build generates the
 parser on Windows

JJTree reads the .jjt under GRAMMAR_ENCODING but writes the intermediate .jj
with the JVM default charset. On a platform whose default is a legacy code
page the grammar's Unicode character classes are written lossily, and JavaCC
then rejects the mangled ranges:

    Error: Line 2331, Column 294: Right end of character range '?' has a
    lower ordinal value than the left end of character range '?'.

The Gradle build has pinned -Dfile.encoding=UTF-8 in gradle.properties since
2021, which is why only the Maven build is affected. Give it the same pin via
.mvn/jvm.config and put windows-latest back in the maven_verify matrix.
---
 .github/workflows/ci.yml | 3 +--
 .mvn/jvm.config          | 1 +
 pom.xml                  | 5 +++++
 3 files changed, 7 insertions(+), 2 deletions(-)
 create mode 100644 .mvn/jvm.config

diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
index 6f2fa19db..a36189576 100644
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -64,8 +64,7 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        # windows-latest disabled: see #<issue> — record the reason, not just the comment
-        os: [ ubuntu-latest, macos-latest ]
+        os: [ ubuntu-latest, windows-latest, macos-latest ]
     steps:
       - uses: actions/checkout@v5
         with:
diff --git a/.mvn/jvm.config b/.mvn/jvm.config
new file mode 100644
index 000000000..7fecadb38
--- /dev/null
+++ b/.mvn/jvm.config
@@ -0,0 +1 @@
+-Dfile.encoding=UTF-8
diff --git a/pom.xml b/pom.xml
index 212e1fe8f..f73fea0f9 100644
--- a/pom.xml
+++ b/pom.xml
@@ -426,6 +426,11 @@
                         <goals>
                             <goal>jjtree-javacc</goal>
                         </goals>
+                        <!--
+                            GRAMMAR_ENCODING governs reading only: JJTree writes the intermediate
+                            .jj with the JVM default charset, which mangles the grammar's non-ASCII
+                            character classes unless file.encoding is UTF-8 (see .mvn/jvm.config).
+                        -->
                         <configuration>
                             <javaccCmdLineArgs>
                                 <arg>-CODE_GENERATOR:"Java"</arg>

From 8c65dabab3733179ab2ff2ddb228d5490d47a3fa Mon Sep 17 00:00:00 2001
From: Knut Wannheden <knut@moderne.io>
Date: Thu, 10 Sep 2026 09:19:42 +0200
Subject: [PATCH 2/2] build: pin file.encoding for the forked test JVMs as well

ParserKeywordsUtilsTest regenerates the .jj in-process and reads it back, so it
depends on the default charset the same way the build does. Surefire forks its
test JVMs and those do not inherit .mvn/jvm.config, leaving them on the platform
default:

    ParserKeywordsUtilsTest.getAllKeywordsUsingJavaCC ... <<< ERROR!
    org.javacc.parser.MetaParseException

Gradle needs no equivalent: its test workers take their encoding from the daemon,
which gradle.properties already pins.
---
 pom.xml | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/pom.xml b/pom.xml
index f73fea0f9..1f28ee1d2 100644
--- a/pom.xml
+++ b/pom.xml
@@ -569,11 +569,13 @@
                         @{jacocoArgLine} is resolved late, after jacoco:prepare-agent has run.
                         Without it the JaCoCo agent would be dropped and coverage reports empty.
                     -->
+                    <!-- Forked test JVMs do not inherit .mvn/jvm.config, so pin the encoding here too. -->
                     <argLine>
                         @{jacocoArgLine}
                         --add-opens=java.base/java.lang=ALL-UNNAMED
                         --add-opens=java.base/java.util=ALL-UNNAMED
                         -Xmx2G -Xms800m -Xss4m
+                        -Dfile.encoding=UTF-8
                     </argLine>
                 </configuration>
             </plugin>