We have an existing `version_key` helper function for sorting versions.
It also does a better job, producing output like:
```
All recent (non-legacy) installable versions are:
3.1.31
3.1.31-asserts
3.1.30
3.1.30-asserts
3.1.29
3.1.29-asserts
```
Rather than:
```
All recent (non-legacy) installable versions are:
3.1.31
3.1.30
3.1.29
3.1.28
3.1.27
```
(with -assert versions listed after 3.1.0)
* Add support for Visual Studio 2022 and migrate to using cmake --build when building on Windows. Leverage the VS2019 MSBuild 'Multi-ToolTask' feature CL_MPCount to saturate project builds properly to 100% CPU usage so building LLVM builds different cpp files in parallel. Clean up some code duplication around Visual Studio support.
* flake
* Work around Linux bot not having 'cmake --build . -j' flag.
Newer versions of emscipten, starting all the way back in 1.39.13, can
automatically locate the `.emscripten` config file that emsdk creates so
there is no need for the explicit EM_CONFIG environment variable. Its
redundant and adds unnessary noisce/complexity.
Really, adding emcc to the PATH is all the is needed these days.
One nice thing about this change is that it allows folks to run
whichever emcc they want to and have it just work, even if they have
configured emsdk. Without this change, if I activate emsdk and I run
`some/other/emcc` then emsdk's `EM_CONFIG` will still be present and
override the configuration embedded in `some/other/emcc`.
e.g. in the same shell, with emsdk activated, I can run both these
commands and have them both just work as expected.
```
$ emcc --version
$ /path/to/my/emcc --version
```
Without this the recommended way to silence emsdk_env was to pipe its
stderr to /dev/null.. but then you also loose potentially useful error
message.
Fixes: #946
* Pass -DENABLE_WERROR=0 when building Binaryen
We are experiencing an issue at our Unity CI with building Binaryen: https://github.com/WebAssembly/binaryen/issues/4588
It seems that for end users, disabling -Werror is a good general measure to enable wider chance of success to build. Emsdk installations are unlikely to be used by Binaryen developers to iterate on Binaryen development, so it is not necessary there?
* Flake
* Enable building LLVM with about-to-be deprecated toolchains for best chance of succeeding with the build.
* Shorten line length
* Adjust to 80 columns
When doing a git clone of a branch, instead of doing a general git clone that first would check out the default (main/master) branch and then later doing a checkout to switch to the target branch, instead specify the `--branch` option to the clone command line to immediately clone and checkout the desired final branch.
This helps first checkout runtime performance on CIs, especially when using shallow clones (GIT_CLONE_SHALLOW option).
Also this sidesteps a really odd git clone issue we are currently seeing in our CI, where it is unable to `checkout` the googletest submodule in binaryen, but would fail saying "path not found".
* Fix native Closure Compiler to work. Reverts my earlier PR #803 which was a great mistake, as it caused Emscripten to silently fall back to Java version of the Closure Compiler. After this PR, Closure Compiler will work on user platforms that do not have Java installed. Also forcibly remove Java version of Closure Compiler on systems where installing the native version succeeds, in order to save on code size.
* Add note to bug
* Improve google-closure-compiler-java uninstall.
* Read Closure version from Emscripten repository
* Skip native google-closure-compiler install when it is not present in the emscripten branch in package.json
* Print error
* Cache git executable search into a variable. This helps reduce noise in the verbose debug logging output messages when EMSDK_VERBOSE=1 is enabled.
* Flake
* Mark cached_git_executable global
Regarding the change to `content_exists`, we only support tools that
are directories full of files, there is no such thing as a tool that
is just a single file and not a directory.
These days this argument really means `install_even_if_directory_exists`
(at least since #9300.
However by the time we call `download_and_unzip` we have already checked
that `is_installed()` is false so we know we want to install for sure.
If the installation directory already existed and contained the correct
contents we would never get as far as `download_and_unzip`.
They were originally removed from emsdk_manifest.json in commit
12f1824ffb ("Remove Tools and SDKs
that do no longer work after migration from Mozilla to Google
hosting. (#395)").
Apparently 10.11 is no longer good enough to run the latest version of
binaryen. Specifically since binaryen switched to using std::variant it
now fails to build with this set to 10.11.
This is also the version used on the emscripten-releases CI which builds
the emsdk binaries:
https://chromium.googlesource.com/emscripten-releases/+/refs/heads/main/src/build.py#673
Previously if a tool (any part of an SDK) was not installed
we would issue a warning and continue to active without returning
non-zero.
This meant doing `emsdk install 2.0.0 && emsdk activate latest`
would appear to be work aside from the warning messages about
latest not being installed.
This is especially annoying since we dropped support for side
by side SDK installations. The following sequence is no longer
valid and we want to make that clear by erroring out:
```
$ emsdk install 2.0.1
$ emsdk install 2.0.2
$ emsdk activate 2.0.1
```
Since 2.0.2 replaces 2.0.1 on the filesystem the active here
could fail hard rather than just warning.
