Hello world does not compile on WSL2 on ReFS filesystem

[ivanstepanovftw]

Zig Version

0.15.0-dev.1228+6dbcc3bd5

Steps to Reproduce and Observed Behavior

Reproducible on Windows 11 with ReFS filesystem.
WSL2 installed with archlinux distro.

Tested on ReFS filesystem listed as type 9p.

$ mount
...
C:\ on /mnt/c type 9p (rw,noatime,aname=drvfs;path=C:\;uid=0;gid=0;symlinkroot=/mnt/,cache=5,access=client,msize=65536,trans=fd,rfd=6,wfd=6)

I haven't tested this on NTFS as I don't have it.

Reproducer:

wsl --install archlinux

cd /mnt/c/Users/User  # <- important
mkdir ziginit
cd ziginit
zig init
zig build

Output:

[root@DESKTOP-P6Q15G3 ziginit]# zig build
error: failed to rename compilation results ('.zig-cache/tmp/1ed140a89cbf7917') into local cache ('.zig-cache/o/e51877309a3ece206e0909199e62559d'): AccessDenied

Expected Behavior

Build is successful.

[mlugg]

WSL is not a real Linux environment, and contains various deviations from the OS proper. These are bugs, and we do not support them. This looks to me like one of those WSL bugs, so I am going to close this issue.

If anyone encounters this issue on a true Linux installation, let me know (with relevant details) and I can re-open this issue.

[Daimanta]

I would ask for a reconsideration on the closing of the ticket. I think the support of WSL as a development platform is useful if the cost of support is not excessive. I would posit three arguments why this issue should be looked at.

1. Using ReFS as a development space has worked for a long time

I stated using Zig at version 0.9, primarily using ReFS as a place to store my code. In that time I never had any issue with any Zig bugs up until version 0.14. This either means that developers put a lot of time into getting this working, or, and this is what I assume, if you support Linux APIs then using ReFS basically works out of the box. I do not think that the argument that ReFS compatilbility is burdensome enough to elide supporting it.

2. WSL usage is quite a popular way of using Linux

WSL was developed many years ago by Microsoft. In the beginning this had a very experimental side but over the years it has developed into a valued part of the Linux ecosystem that is supported by Microsoft. Several companies provide support of Linux by providing a Windows machine with WSL instead of dual boot or VMs. In the company I work at, this is the primary method of using Linux and it is essential infrastructure for development. Therefore I do not think that ignoring WSL is a good idea if you want to include popular ways of using Zig. When Zig becomes more stable and more production ready, I would assume that the segment of people who want to use WSL as an environment for development with Zig will go up substantially

3. The issue of this bug seems quite localized and solvable

The error message that is presented can be found in exactly one place in the code. That makes localizing the issue very doable. Furthermore, the issue occurs at around the last possible step in the compilation process. This would indicate that the issue on WSL doesn't affect an fundamental part of the code and would be solvable upon a minor inspection of the code. If there are many developers who use WSL (with or without ReFS), it would be considered a great help if this issue would be fixed. If need be, I would be willing to look at the causes of the issue if you are willing to consider fixing this issue.

[Khitiara]

Seconding the request that this be reconsidered - primarily because

1. the issue is specifically a regression, and
2. zig fetch has now been broken twice on windows which cannot be worked around because of this issue

potentially relevant information: microsoft/WSL#12897

[Khitiara]

pinned the issue down to be with the self-hostex x86_64 backend specifically