bug#49990: [core-updates-frozen] Ant-bootstrap broken by classpath-bootstrap

["Ludovic Courtès" <ludo@gnu.org>]

[-- Attachment #1: Type: text/plain, Size: 2134 bytes --]

Hello!

As discussed on IRC, I had a semi-victory with the attached patch, which
works around a miscompilation issue in ‘Java_java_io_VMFile_isFile’.

Unfortunately, with this patch applied, ‘ant-bootstrap’ fails to build
with:

--8<---------------cut here---------------start------------->8---
$ ./pre-inst-env guix build -e '(@@ (gnu packages java) ant-bootstrap)' -K
[…]
... Copying Required Files
... Building Ant Distribution
Buildfile: /tmp/guix-build-ant-bootstrap-1.8.4.drv-0/apache-ant-1.8.4/build.xml

BUILD FAILED
Could not load the version information.

Total time: 0 seconds
... Failed Building Ant Distribution !
error: in phase 'build': uncaught exception:
%exception #<&invoke-error program: "bash" arguments: ("bootstrap.sh" "-Ddist.dir=/gnu/store/88qc3rkp3bc6qsf6gmknv51vprd3r8j4-ant-bootstrap-1.8.4") exit-status: 1 term-signal: #f stop-signal: #f> 
phase `build' failed after 0.8 seconds
command "bash" "bootstrap.sh" "-Ddist.dir=/gnu/store/88qc3rkp3bc6qsf6gmknv51vprd3r8j4-ant-bootstrap-1.8.4" failed with status 1
--8<---------------cut here---------------end--------------->8---

The message “Could not load the version information.” indicates a
NullPointerException in Main.java:

--8<---------------cut here---------------start------------->8---
      try {
          Properties props = new Properties();
          InputStream in =
              Main.class.getResourceAsStream("/org/apache/tools/ant/version.txt");
          props.load(in);
          // …
      } catch (NullPointerException npe) {
          throw new BuildException("Could not load the version information.");
      }
--8<---------------cut here---------------end--------------->8---

Specifically, ‘in’ is null.

‘version.txt’ is looked for in
/tmp/guix-build-ant-bootstrap-1.8.4.drv-0/apache-ant-1.8.4/build/classes/org/apache/tools/ant/version.txt
but at this point it’s still in src/ only.  My understanding is that the
“build” target in ‘build.xml’ should copy it to build/classes/.

Ideas?  What a wonderful puzzle we have!  :-)

Thanks,
Ludo’.


[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: Type: text/x-patch, Size: 2874 bytes --]

diff --git a/gnu/packages/java.scm b/gnu/packages/java.scm
index 08ef7a8213..d3c95a456d 100644
--- a/gnu/packages/java.scm
+++ b/gnu/packages/java.scm
@@ -230,7 +230,8 @@ only faster.")
               (sha256
                (base32
                 "0i99wf9xd3hw1sj2sazychb9prx8nadxh2clgvk3zlmb28v0jbfz"))
-              (patches (search-patches "classpath-aarch64-support.patch"))))
+              (patches (search-patches "classpath-aarch64-support.patch"
+                                       "classpath-miscompilation.patch"))))
     (build-system gnu-build-system)
     (arguments
      `(#:configure-flags
@@ -247,17 +248,6 @@ only faster.")
              "--disable-gjdoc")
        #:phases
        (modify-phases %standard-phases
-         ;; XXX: This introduces a memory leak as we remove a call to free up
-         ;; memory for the file name string.  This was necessary because of a
-         ;; runtime error that would have prevented us from building
-         ;; ant-bootstrap later.  See https://issues.guix.gnu.org/issue/36685
-         ;; for the gnarly details.
-         (add-after 'unpack 'remove-call-to-free
-           (lambda _
-             (substitute* "native/jni/java-io/java_io_VMFile.c"
-               (("result = cpio_isFileExists.*" m)
-                (string-append m "\n//")))
-             #t))
          (add-after 'install 'install-data
            (lambda _ (invoke "make" "install-data"))))))
     (native-inputs
diff --git a/gnu/packages/patches/classpath-miscompilation.patch b/gnu/packages/patches/classpath-miscompilation.patch
new file mode 100644
index 0000000000..835113df71
--- /dev/null
+++ b/gnu/packages/patches/classpath-miscompilation.patch
@@ -0,0 +1,31 @@
+For some reason, the original code gets miscompiled on x86_64, leading
+'Java_java_io_VMFile_isFile' to return true when the return value of
+'cpio_checkType' is ENOENT (= 2).
+
+See <https://issues.guix.gnu.org/issue/36685>
+and <https://issues.guix.gnu.org/49990>.
+
+diff --git a/native/jni/java-io/java_io_VMFile.c b/native/jni/java-io/java_io_VMFile.c
+index de1320b..9a5d375 100644
+--- a/native/jni/java-io/java_io_VMFile.c
++++ b/native/jni/java-io/java_io_VMFile.c
+@@ -278,6 +278,7 @@ Java_java_io_VMFile_isFile (JNIEnv * env,
+   const char *filename;
+   int result;
+   jint entryType;
++  jboolean isfile;
+ 
+   /* Don't use the JCL convert function because it throws an exception
+      on failure */
+@@ -288,9 +289,10 @@ Java_java_io_VMFile_isFile (JNIEnv * env,
+     }
+ 
+   result = cpio_checkType (filename, &entryType);
++  isfile = (result == CPNATIVE_OK && entryType == CPFILE_FILE ? 1 : 0);
+   (*env)->ReleaseStringUTFChars (env, name, filename);
+ 
+-  return result == CPNATIVE_OK && entryType == CPFILE_FILE ? 1 : 0;
++  return isfile;
+ #else /* not WITHOUT_FILESYSTEM */
+   return 0;
+ #endif /* not WITHOUT_FILESYSTEM */