summarylogtreecommitdiffstats
AgeCommit message (Collapse)Author
2023-04-26Update to 22.3.2Lucas Werkmeister
2023-01-26Update to 22.3.1Lucas Werkmeister
2022-10-25Update to 22.3.0Lucas Werkmeister
2022-07-26Update to 22.2.0Lucas Werkmeister
With optional dependencies on three new packages for components that are no longer included in this package. (graaljs and graal-llvm are required by test.sh; I don’t see a need to explicitly test graal-visualvm there.)
2022-04-26Update to 22.1.0Lucas Werkmeister
2022-01-25Update to 22.0.0.2Lucas Werkmeister
2021-10-19Update to 21.3.0Lucas Werkmeister
2021-07-31Install alpm hook recommending `gu rebuild libpolyglot`Lucas Werkmeister
Apparently the only reason that the `node --polyglot` added to test.sh in the previous commit worked on my system is that it silently fell back to `node --polyglot --jvm`; however, it’s *supposed* to run polyglot Node + everything else as a native image, without a JVM. And apparently, R can’t be built as a native image (for engineering reasons, I’ve been told), so if it doesn’t fall back to --jvm mode, then our test doesn’t work. Consequently, we should actually specify --polyglot --jvm. The fallback happens when libpolyglot.so is missing. It’s supposed to be built by `gu rebuild libpolyglot`, and apparently `gu` prints a message recommending that you run this command when you use it to install a new language; since we bypass gu, we should print a similar message in our package installation process. We don’t directly run the command for the same reason that gu doesn’t: it’s way too expensive, requiring over four hours of CPU time on my system. Even as a hook, rather than a post-install script (i.e. only running once at the end of all package installs, not once per package), I can’t justify this expenditure for everyone installing the package, especially since most people probably won’t need this (evidently I’ve been using GraalVM for years without ever having built libpolyglot.so). So we just make the hook print a message for users, and they can hopefully decide for themselves if they want to run the command or not. (If people ask about the message in the AUR comments, we can probably iterate on the phrasing a bit.)
2021-07-31Use Node.js launcher in test.shLucas Werkmeister
The Node.js component is no longer included in this package, making it separate from the JS language (which is still included), so we should make sure to test it in test.sh. The simplest way to do that is to replace R with Node as the host language for the polyglot test.
2021-07-20Update to 21.2.0Lucas Werkmeister
2021-04-22Fix optdepends nodejs JDK versionLucas Werkmeister
2021-04-20Recommend graal-nodejs-jdk11-bin packageLucas Werkmeister
The GraalVM 21.0.0 base download does not include the Node.js component anymore, so that’s available as a separate package now. Add it as an optional dependency of this one, so that people can find it easily (since it used to be bundled with this package).
2021-04-20Add test for Espresso (Java on Truffle)Lucas Werkmeister
I don’t think we can test interop with other languages easily (Espresso source is JVM bytecode, not Java source code), but we can at least test it as the main language, similar to native-image.
2021-04-20Update to 21.1.0Lucas Werkmeister
2021-02-14Update to 21.0.0.2Lucas Werkmeister
2021-01-19Update to 21.0.0Lucas Werkmeister
And silence emcc in test.sh, since it otherwise prints some noise to stderr when some compilation parts aren’t cached.
2020-11-17Add GraalWasm to test.sh scriptLucas Werkmeister
Needs Emscripten to compile the C file, which apparently puts the emcc binary in an unusual location, so add that to the PATH as well. I also considered adding the .wasm file to the Git repository directly, but it seems the format changes occasionally – a file I used to test earlier GraalWasm releases didn’t work with 20.3.0 anymore – so let’s instead have the test script compile the source code from scratch, and require the testing user to have emscripten installed.
2020-11-17Update to 20.3.0Lucas Werkmeister
2020-08-19Update to 20.2.0Lucas Werkmeister
2020-05-22Update to 20.1.0Lucas Werkmeister
2020-02-19Update to 20.0.0Lucas Werkmeister
2020-01-28Add AArch64 versionLucas Werkmeister
The corresponding files have been available upstream for a few releases already, I just never thought to make them available. Thanks to AUR user alex4o [1] for the tip! [1]: https://aur.archlinux.org/account/alex4o/
2020-01-15Update to 19.3.1Lucas Werkmeister
2020-01-04Update to 19.3.0.2Lucas Werkmeister
2019-12-04Don’t remove static librariesLucas Werkmeister
If we don’t specify the 'staticlibs' option, makepkg will remove static library (.a) files when building the package, which breaks native-image. Many thanks to AUR user mcescher [1] for figuring this out! [1]: https://aur.archlinux.org/account/mcescher/
2019-11-25Initial commitLucas Werkmeister
With the 19.3.0 release, GraalVM supports two Java versions: Java 8 and Java 11. Accordingly, the former graal-bin package is split into two: jdk8-graalvm-bin and jdk11-graalvm-bin (this package). We use this opportunity to make several other adjustments as well. The main product name seems to be “GraalVM”, not just “Graal”, so we reflect this in the package name and in the Java name. We no longer provide a version of the package name without the “-bin” suffix. It doesn’t seem likely that a built-from-source version will be added to the AUR soon, and other packages like jdk8-openj9-bin or jdk8-j9-bin don’t provide non-bin versions either. We no longer recommend a native-image package. The old package only did this because it used to include that functionality; this new package has never included native-image, so I don’t see the need to recommend it. People interested in it should be able to find it easily enough. A test script is included, which I’d been using locally for some time already. It assumes you’ve locally installed current versions of the GraalVM, FastR, TruffleRuby, GraalPython and native-image packages, and is typically run just before pushing the updated PKGBUILDs to the AUR. Note that the jdk11 version of native-image is not yet functional, so the script currently fails. And finally, a .gitignore file never hurts.