[jdk11u-dev] RFR: 8357657: [11u][windows] cannot stat '/jdk.crypto.ec/*': No such file or directory

[Aleksey Shipilev]

On Fri, 23 May 2025 16:12:57 GMT, Sergey Bylokhov <serb at openjdk.org> wrote:

> This patch provides a workaround for a bug in the latest version of Cygwin. Only JDK 11 is affected; newer versions are not impacted because the makefiles were updated by [JDK-8220383](https://bugs.openjdk.org/browse/JDK-8220383).
> 
> The issue is that:
> `ifneq ($(wildcard $(IMPORT_MODULES_CLASSES)/$(MODULE)), )`
> returns results even when `IMPORT_MODULES_CLASSES` is empty/unset. This causes the build to incorrectly search for a module like `/jdk.crypto.ec/`, which fails because the directory does not exist.
> 
> The fix is to first check whether `IMPORT_MODULES_CLASSES` is set. If it is, we define `IMPORT_MODULE_DIR` as `$(IMPORT_MODULES_CLASSES)/$(MODULE)` and use this property consistently throughout the build. This matches the approach introduced in [JDK-8220383](https://cr.openjdk.org/~erikj/8220383/webrev.01/make/CompileJavaModules.gmk.sdiff.html) for this codepath.

Marked as reviewed by shade (Reviewer).

OK, so this effectively takes the first two hunks from https://github.com/openjdk/jdk/commit/2af935c800691492f3e84cc6a11c4ec8f4070dba. I think it is reasonable to pull it in selectively into 11u. Have you considered backporting [JDK-8220383](https://cr.openjdk.org/~erikj/8220383/webrev.01/make/CompileJavaModules.gmk.sdiff.html) wholesale? I am guessing it is more risky?

-------------

PR Review: https://git.openjdk.org/jdk11u-dev/pull/3035#pullrequestreview-3442194759
PR Comment: https://git.openjdk.org/jdk11u-dev/pull/3035#issuecomment-3510631370