Tracking issue for Windows ARM64 support

[dennisameling]

There's been a discussion about Windows ARM64 support here: #3021

I made a discussion comment there to keep track of the outstanding tasks to get basic ARM64 support to Git for Windows. However, @Alovchin91 informed me via Twitter that it was hard to keep track of the progress that way, requesting a dedicated issue. So here it is 😊

Open tasks for basic ARM64 support:

• ✔️ Enable compiling Git for Windows (git.exe etc.) natively for arm64 Windows arm64 support #2915 (comment)
• ✔️ Fix double .exe suffix in CMake generated build Fix double .exe suffixes in CMakeLists.txt  #2966
• ✔️ Include vcpkg DLLs in CMake build cmake(install): include vcpkg dlls #2971
• ✔️ Add ARM64 to /etc/profile added MSYSTEM=ARM64 case in /etc/profile git-sdk-32#6
• ✔️ Update /etc/profile in git-extra.install.in Start support for MSYSTEM=ARM64 build-extra#321
• ✔️ Modify git-wrapper in mingw32 to use the native git.exe, set MSYSTEM=ARM64, etc. Add arm64 support to git-wrapper MINGW-packages#44
• ✔️ Create portable version of ARM64 Git for Windows portable: add arm64 artifact option build-extra#323
• ✔️ Create MinGit package for ARM64 Git for Windows mingit: add arm64 artifact option build-extra#324
• ✔️ Create archive package for ARM64 Git for Windows archive: add arm64 artifact option build-extra#325
• ✔️ Create an installer for ARM64 installer: add arm64 artifact option build-extra#326
• ✔️ Add ARM64 artifacts to CI git-artifacts: add ARM64 artifacts #3017
• ✔️ Update Inno Setup installer in preparation for installing Git for Windows ARM64 to the 64-bit Program Files folder installer: Update Inno Setup from 5.6.1 to 6.1.2 build-extra#330
• ✔️ Create logic for ARM64 in Inno Setup installer, so that Git will be installed in the 64-bit Program Files folder installer: add 64-bit support for ARM64 build-extra#333
• ✔️ Fix Git Credential Manager Core (necessary for git push etc.) by referencing it from the mingw32 directory Git Credential Manager Core missing in ARM64 libexec folder #3015
• ✔️ Decrease the size of the ARM64 artifacts significantly Assorted fixes/enhancements for the git-artifacts workflow #3053
• ✔️ Fix issue where system-wide gitconfig can't be read ARM64: cannot read/write system gitconfig  #3071
• ✔️ Fix bug where cmd\git.exe is crashing when arm64/bin folder is present cmd\git.exe crashing when ARM64 folder is present in Git installation (git-wrapper) #3083
• ✔️ Fix issue with Git Credential Manager Core on ARM64 git-artifacts: Update ARM64 GCM Core workaround #3085
• ❓ Update Azure DevOps pipeline?
• ✔️ Update the daily "Git for Windows update check" for ARM64: git update-git-for-windows: download ARM64 Git for Windows if applicable build-extra#340
• ✔️ git --version --build-options returns AMD64 as the CPU cmake(): allow setting HOST_CPU for cross-compilation #3327
• Visual C++ for ARM64 needs to be installed, due to us using MSVC for the ARM64 builds. We can either include that in the installer or statically build (if possible)
• Documentation using --help (example: git log --help) doesn't work since Git looks for docs in the wrong location

Thanks to the changes that we did in the last few days, there's now a fully working build of Git for Windows ARM64.

Test build for ARM64 users, built on March 12, 2021: https://github.com/dennisameling/git/releases/tag/v2.31.0-rc2.windows.2

I think we're pretty much ready to publish a beta version of Git for Windows ARM64! @dscho what do you think? Please let me know if I missed something in the list above.

[dscho]

I'm a bit focused on making sure that the v2.31.0 release goes smoothly, and will be happy to pay more attention to the ARM64 side of things after that release is out (which will probably happen Tuesday or Wednesday this coming week).

[Alovchin91]

Hi @dennisameling ,

Thanks for your wonderful work! ❤️ Have been using MinGit lately (which is a 💎 by itself) and it works great on my SPX!

One thing I just found:

C:\Projects\rustup>git log --help
fatal: 'C:/MinGit/arm64/share/doc/git-doc': not a documentation directory.

While MinGit doesn't seem to contain git-doc anyway, what caught my attention is that git.exe tries to search for documentation in the arm64 directory, while docs seem to still be in mingw32.

Could you please check on your side, on a complete Git setup? Thanks a lot! 😃

[dscho]

The documentation is not built as part of the CMake-based build. It will take quite some work to implement that. @Alovchin91 that's your chance to get involved! 😉

[Alovchin91]

@dscho I would be happy to help, but right now my capacity is extremely limited unfortunately and I have never had any proper experience with Cmake.

But wouldn't it be enough to point Git.exe from arm64 folder to the documentation in mingw32? 🤔

[dscho]

But wouldn't it be enough to point Git.exe from arm64 folder to the documentation in mingw32? 🤔

Not really, because the build configuration may differ in more than just the CPU architecture, even in ways affecting how/what parts of the documentation are compiled.

[Alovchin91]

Hmm, I think right now my concern is mostly about where does arm64 version look for docs in a full installer-based version of Git 🤔

I saw that it's also absent from the "normal" mingw32 MinGit build, so I suppose missing docs in those builds is not arm64 specific issue.

[dscho]

Right, documentation is excluded from MinGit builds.

But looking for the documentation in the wrong location, just because there is no documentation in the correct location, isn't a solution for the non-MinGit flavors of Git for Windows, either ;-)

[Alovchin91]

Ah, right, now I get your point 😁 For some reason I thought that the most part of the tools are redirected to mingw32 folder and arm64 only contains wrappers or helpers 😅

[dscho]

Thanks to @dennisameling tireless work, the Git executables are built for Windows/ARM64. The i686 version of the MINGW tools we're still relying on are things like curl.exe, but we do include all of the i686 Git executables at the moment, too.

[dennisameling]

Can confirm the help docs are only present in mingw32\share\doc\git-doc on a full installation and not in arm64\share\doc\git-doc.

The help function looks at GIT_HTML_PATH to find the help docs:

git/builtin/help.c

Lines 464 to 482 in 959b2f0

	static void get_html_page_path(struct strbuf *page_path, const char *page)
	{
	struct stat st;
	char *to_free = NULL;
	if (!html_path)
	html_path = to_free = system_path(GIT_HTML_PATH);
	/* Check that we have a git documentation directory. */
	if (!strstr(html_path, "://")) {
	if (stat(mkpath("%s/git.html", html_path), &st)
	|| !S_ISREG(st.st_mode))
	die("'%s': not a documentation directory.", html_path);
	}
	strbuf_init(page_path, 0);
	strbuf_addf(page_path, "%s/%s.html", html_path, page);
	free(to_free);
	}

This is defined in the CMake file:

git/contrib/buildsystems/CMakeLists.txt

Lines 206 to 217 in 959b2f0

	add_compile_definitions(PAGER_ENV="LESS=FRX LV=-c"
	GIT_EXEC_PATH="libexec/git-core"
	GIT_LOCALE_PATH="share/locale"
	GIT_MAN_PATH="share/man"
	GIT_INFO_PATH="share/info"
	GIT_HTML_PATH="share/doc/git-doc"
	DEFAULT_HELP_FORMAT="html"
	DEFAULT_GIT_TEMPLATE_DIR="share/git-core/templates"
	GIT_VERSION="${PROJECT_VERSION}.GIT"
	GIT_USER_AGENT="git/${PROJECT_VERSION}.GIT"
	BINDIR="bin"
	GIT_BUILT_FROM_COMMIT="")

@dscho would you be okay with a clause in the CMake file for Win ARM64 that sets GIT_HTML_PATH to ../mingw32/share/doc/git-doc for the time being? We can leverage the already available CMAKE_GENERATOR_PLATFORM for that logic.

[dscho]

would you be okay with a clause in the CMake file for Win ARM64 that sets GIT_HTML_PATH to ../mingw32/share/doc/git-doc for the time being?

I wonder how much work it would be to optionally use asciidoctor or asciidoc, if configured. These lines describe how asciidoc is called: https://github.com/git/git/blob/v2.31.0/Documentation/Makefile#L125-L133 And these describe how asciidoctor would be called instead: https://github.com/git/git/blob/v2.31.0/Documentation/Makefile#L179-L190

[dennisameling]

I wonder how much work it would be to optionally use asciidoctor or asciidoc, if configured.

Will have a look in the coming days.

Just created a PR for 64-bit ARM64 installer support: git-for-windows/build-extra#333

[dennisameling]

@dscho now that the 64-bit ARM64 installer PR has been merged, do you think we're ready to release a beta build of Git for Windows ARM64? I'm happy to look into the docs thing with asciidoctor and asciidoc mentioned above, but will only have time for that after May 7 due to work responsibilities and a break 😊 Thanks in advance!

[rimrul]

Check if the daily "Git for Windows update check" works correctly on ARM64

It'll probably run daily and check for updates, but when it detects an update, it'll try to download and install the x86 installer.

[dscho]

do you think we're ready to release a beta build of Git for Windows ARM64?

Yes, I think we're getting to the point where we want to invite testers (I, unfortunately, am not eligible because I lack the hardware).

I'm happy to look into the docs thing with asciidoctor and asciidoc mentioned above

Fantastic.

I do think that it should be possible to add that to the CMakeLists.txt definition (a quick search brought up gdamore/nanomsg@5592124 which might be a good example to follow, although I would try not to have a separate Boolean to enable documentation; Rather, I would make it conditional on a check that tries to find asciidoc or asciidoctor, allowing the user to specify a hard-coded path to either to side-step the discovery).

It'll probably run daily and check for updates, but when it detects an update, it'll try to download and install the x86 installer.

Indeed.

For microsoft/git, I already did something slightly ugly to redirect the git update-git-for-windows command: microsoft#321 (comment). Essentially, it is a hack that edits the git-update-git-for-windows script to target a different update site, via a few sed commands.

I could imagine that we would probably want to improve on that e.g. by enhancing git-update-git-for-windows itself. One option would be to search for a file, say, git-update-git-for-windows.ini next to the script (when called via the PATH, as git update-git-for-windows will do internally, "$0" contains the absolute path to the script, therefore "$0.ini"would refer to that.inifile's path). If found, it would override a couple of things. Something along these lines (applies tobuild-extra`, but reader beware: this is totally untested):

diff --git a/git-extra/git-update-git-for-windows b/git-extra/git-update-git-for-windows
index 7121dd7be..50ff5524a 100755
--- a/git-extra/git-update-git-for-windows
+++ b/git-extra/git-update-git-for-windows
@@ -174,7 +174,22 @@ update_git_for_windows () {
 		;;
 	esac
 
-	latest_tag_url=https://gitforwindows.org/latest-tag.txt
+	if test -f "$0.ini"
+	then
+		fork="$(git config -f "$0.ini" update.fromFork)"
+		test -n "$releases_url" || {
+			echo "Could not find update.fromFork in $0.ini" >&2
+			return 1
+		}
+		releases_url=https://api.github.com/repos/$fork/releases
+		latest_tag_url=$releases_url/latest
+		latest_eval='latest=${latest_tag#*\"tag_name\": \"}; latest=${latest%%\"*}'
+	else
+		releases_url=https://api.github.com/repos/git-for-windows/git/releases
+		latest_tag_url=https://gitforwindows.org/latest-tag.txt
+		latest_eval='latest=${latest_tag#v}'
+	fi
+
 	latest_tag=$(http_get $latest_tag_url) ||
 	case $?,"$proxy" in
 	7,)
@@ -190,7 +205,7 @@ update_git_for_windows () {
 		;;
 	esac
 
-	latest=${latest_tag#v}
+	eval "$latest_eval"
 	# Did we ask about this version already?
 	recently_seen="$(get_recently_seen)"
 	test -n "$quiet" && test "x$recently_seen" = "x$latest" && return
@@ -215,7 +230,6 @@ update_git_for_windows () {
 	esac
 
 	echo "Update $latest is available" >&2
-	releases_url=https://api.github.com/repos/git-for-windows/git/releases
 	releases=$(http_get $releases_url/latest) || return
 	download=$(echo "$releases" |
 		grep '"browser_download_url": "' |

[dscho]

❓ Check if the daily "Git for Windows update check" works correctly on ARM64

Right, completely forgot to address that part. We do trust uname -m in the git-update-git-for-windows script, but on ARM, we cannot (because our MSYS2 will be an i686 or an x86_64 one).

[rimrul]

I could imagine that we would probably want to improve on that e.g. by enhancing git-update-git-for-windows itself. One option would be to search for a file, say, git-update-git-for-windows.ini next to the script (when called via the PATH, as git update-git-for-windows will do internally, "$0" contains the absolute path to the script, therefore "$0.ini"would refer to that.ini` file's path). If found, it would override a couple of things.

I was thinking a lot simpler: We could check $MSYSTEM or $MINGW_MOUNT_POINT to detect we're on an ARM install. Admittedly, I did not think about microsoft/git when I had that idea.

[dscho]

I don't really trust MSYSTEM and MINGW_MOUNT_POINT at this stage ;-) If git.exe sets MSYSTEM, for example, it will not have the expected value.

[dscho]

I opened git-for-windows/build-extra#338 to help with this.

[dennisameling]

Incredibly excited that git-for-windows/build-extra#340 has been merged. I almost don't dare to ask, but do you think we're now ready to release a beta version of Git for Windows ARM64, @dscho (apart from the request for the docs, which I hope to look into in the coming two weeks)? 🤞🏼

Yes, I think we're getting to the point where we want to invite testers (I, unfortunately, am not eligible because I lack the hardware).

I ordered and received a Raspberry Pi 4 last week, then installed Windows on ARM 19043 (21H1) on it. The performance is surprisingly good, but every once in a while there's a little driver-related hiccup that the Windows on Raspberry team is working hard on. If you want, I can send you the TeamViewer ID + password so you can test a bit on your end as well 👍🏼

[Alovchin91]

I have the Surface Pro X so I can also help with testing 🙂 RDP is also an option.

[dscho]

Right now, I would like to focus on Git for Windows v2.32.0, for which -rc0 came out yesterday, and the final version is expected around June 7th/8th.

After that, I'm all game for ARM64 and will take y'all up on the kind offers to let me connect to your hardware.

[Alovchin91]

Well... I wonder if BUILD next week brings anything interesting to arm64 world anyway 🤷 Maybe an arm64 GitHub Actions pipeline? One could dream...

[rimrul]

@dennisameling Could you add the issue with git --version --build-options that I pointed out in #3083 to the list?

While it isn't a show stopper for the ARM beta it would be nice to have that fixed before we get more ARM bug reports.

[Alovchin91]

Okay, I hope it's finally time to prepare the first ARM64 beta! 🤗 Please let me know if I can help with anything. I could provide an ARM64 virtual machine on my Surface Pro X for a build pipeline, for example 🙂

[dennisameling]

Just updated the list above with open issues. I recently came across the fact that Visual C++ is needed since we use MSVC for the ARM64 builds. We can either include the Visual C++ Redistributable package in the installer, or statically build using MSVC (if possible). Do you have any preference here @dscho? I guess the latter option would be best as software like GitHub Desktop is depending on MinGit, so a static build would make their lives easier as well.

[rimrul]

I don't think UCRT can be statically linked and redistributing vcredist would probably be a GPL violation, so linking the vcredist download page is probably our only option.

[dscho]

I am still a bit uneasy about the double-fork deadlock, even though there seems to be some movement on about a potential work-around.

This work-around has now been contributed to the cygwin-patches mailing list. I have updated git-for-windows/msys2-runtime#73 and am currently testing. So far, everything works!

Therefore you can add @jeremyd2019 to your heroes, @Alovchin91!

[Alovchin91]

Therefore you can add @jeremyd2019 to your heroes, @Alovchin91!

Jeremy has been in my list of heroes for quite some time now 😉

Nice to see this workaround making its way into Git for Windows! I wonder though if Cygwin will be eager to merge it, or they'd rather wait for the ARM64 support in GCC to arrive to compile the whole thing to ARM64. In any case, hopefully this works for Git for now 🤞

BTW I did get some hangs even with the older version of Pacman a few times, but so far that has never happened with Git itself. And I have it set up with SSH commit signing, and have been using Git ARM64 from this repo for about a year now I think.

[shanselman]

AMAZING! Thanks!

[Alovchin91]

Okay, looks like it's time to update my setup to msys2-runtime 3.5.4-4 and unblock Pacman upgrades. I got hangs on newer Pacman versions most of the time, so should be pretty straightforward to verify 😄

[Alovchin91]

...aaand it works! Previously, running /update-via-pacman.bat, I would get hangs like 90% of the time. Now I ran it a number of times with the newest Pacman and it never hung, with no phantom processes staying behind either 👍

An official ARM64 release today? 😜 ('cause everybody knows that Friday is the best day for releases)

[dscho]

running /update-via-pacman.bat, I would get hangs like 90% of the time.

Could you please re-test with /update-via-pacman.ps1? The .bat file is still there for just one reason: time (lack thereof, to be precise).

An official ARM64 release today? 😜 ('cause everybody knows that Friday is the best day for releases)

Alas, I won't be able to do that. Hopefully I'll get around to it next week.

[Alovchin91]

Could you please re-test with /update-via-pacman.ps1? The .bat file is still there for just one reason: time (lack thereof, to be precise).

Hmm, I'm using MSYS2 + the "MSYS2 proper" method, so I don't have that script. Does it need anything additional, or I can just copy it over from the SDK repo to test?

Alas, I won't be able to do that. Hopefully I'll get around to it next week.

No probs, it was a joke 😅 I expected it to be released with the next GfW update tbh.

[Alovchin91]

I've copied it over from the Arm64 SDK repo. It ran without an issue, though it did install the asciidoctor package which, I believe, is only needed for development, and when re-installing git-extra it has shown a warning:

/tmp/alpm_KB3vFO/.INSTALL: line 6: : No such file or directory

I believe I've seen it before with the .bat script when the environment didn't contain a path to git.exe (I think).

I don't get the same warning when running the .bat version.

EDIT 1:

If I understand it correctly, this essentially boils down to this file and line:

https://github.com/git-for-windows/build-extra/blob/5d189481c541e1a9af47d20592d6fb7d4b1e19aa/git-extra/git-extra.install#L6

Running git -c core.editor=echo config --system -e 2>/dev/null gives me this: C:/msys64/clangarm64/etc/gitconfig. I suppose it could be something in the way the path is converted when Pacman is run from inside of PowerShell?

If that's of any importance, I ran the .ps1 script like this: powershell -ExecutionPolicy Bypass /update-via-pacman.ps1

EDIT 2:

Removing the line $env:MSYS2_PATH_TYPE = "minimal" removes the warning.

EDIT 3:

Please correct me if I'm wrong, but it seems like it makes no sense to set $env:PATH = "$(Get-Location)\usr\bin;" + $env:PATH and then set $env:MSYS2_PATH_TYPE = "minimal" and run bash -lc. I guess what is meant here is to set $env:MSYS2_PATH_TYPE to "inherit"?

[dscho]

/add relnote blurb This release comes with the first early native support of Windows/ARM64, ready for testing. Please report any issues!

The workflow run was started

[dscho]

Here it is, Git for Windows/ARM64 v2.47.1: https://github.com/git-for-windows/git/releases

[...]

• Git-2.47.1-arm64.exe
• PortableGit-2.47.1-arm64.7z.exe

On this high note, I am incredibly happy to close this here ticket as addressed!

[aharpervc]

I'm just an outside observer here, however I have been keeping and eye on this issue from the beginning, so it's a big moment -- congrats to everyone who put in the labor on this!

[dscho]

And here is the celebratory BlueSky post.