Github user srowen commented on a diff in the pull request:
https://github.com/apache/spark/pull/17472#discussion_r108884076
--- Diff: common/unsafe/src/main/java/org/apache/spark/unsafe/Platform.java
---
@@ -46,18 +46,22 @@
private static final boolean unaligned;
static {
boolean _unaligned;
- // use reflection to access unaligned field
- try {
- Class<?> bitsClass =
- Class.forName("java.nio.Bits", false,
ClassLoader.getSystemClassLoader());
- Method unalignedMethod = bitsClass.getDeclaredMethod("unaligned");
- unalignedMethod.setAccessible(true);
- _unaligned = Boolean.TRUE.equals(unalignedMethod.invoke(null));
- } catch (Throwable t) {
- // We at least know x86 and x64 support unaligned access.
- String arch = System.getProperty("os.arch", "");
- //noinspection DynamicRegexReplaceableByCompiledPattern
- _unaligned =
arch.matches("^(i[3-6]86|x86(_64)?|x64|amd64|aarch64)$");
+ String arch = System.getProperty("os.arch", "");
+ if (arch.matches("^(ppc64le | ppc64)$")) {
--- End diff --
Sorry, I'm writing one thing and thinking another. I mean alignment, not
endianness. Which architectures do you know allow unaligned access? I'd presume
all PPC does, and I assume the JDK issue means "PPC64 (big-endian) but also
PPC64 little-endian". OK, to be conservative, maybe just check the strings
"ppc64" and "ppc64le" as you intended.
However your regex doesn't work. You have extra whitespace. Just instead
check `arch.equals(...) || arch.equals(...)`
Also the PR title shoudl match the JIRA title. I was commenting that it
differed here in referring to a flaky test, but it isn't. To be very clear, I
propose you rename both to perhaps: "Workaround JDK-8165231 to identify PPC64
architectures as supporting unaligned access"
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
