Sys.detectwsl()

[mgautam98]

Tries to solve #36354

Since it will be a runtime function I did it with Julia itself.

I looked into libuv docs and there is no wrapper function to get proc-version since Its too system specific and libuv abstracts it.

Looking at microsoft/WSL#4071 and javascript/wsl seems that we can only find it from version information

If its not the correct way then please give me some pointers in the correct direction.

[DilumAluthge]

Should there be a fallback in case an error occurs while trying to open and/or read from /proc/version?

[mgautam98]

Should there be a fallback in case an error occurs while trying to open and/or read from /proc/version?

@DilumAluthge Yes I should wrap it around try-catch

[StefanKarpinski]

How about this:

detectwsl(os::Symbol) =
    islinux(os) &&
    isfile("/proc/version") &&
    contains(read("/proc/version", String), r"Microsoft"i)

[StefanKarpinski]

I guess the try-catch might still be good but the main thing is just checking that the file exists.

[StefanKarpinski]

And actually, if the file exists but we can't read it, what value should be returned? An error might be best. That's a pretty weird situation and if you can't open and read the file, how do you know if you're on WSL or not?

[DilumAluthge]

Yeah. Is there ever a non-pathological situation in which /proc/version exists but is not world-readable?

[StefanKarpinski]

I did just try changing the permissions on a Linux system and you can if you have root access change them, so it's possible that someone could make this file not readable to some users. But in that case what's the right answer here? One should generally assume not WSL as the norm, but if you can't read that file we don't know.

[DilumAluthge]

Does Sys.detectwsl() have to definitely return a Bool? Can it instead return Union{Bool, Missing}? Then, if e.g. we can't read /proc/version, we return missing?

[DilumAluthge]

(Could replace Missing with Nothing, if you want.)

[StefanKarpinski]

I feel like that's just deferring the error and making it harder to debug. The inevitable usage of this is to test it with a conditional so that's converting a clear "we cannot tell if this system is running WSL or not" error into an obscure " TypeError: non-boolean (Missing) used in boolean context" error.

[DilumAluthge]

Hmm. Maybe then throw an error if we can't read the file? At least it will then be clear what the source of the error is.

[StefanKarpinski]

Yes, that's why my version just checks if the file exists and then reads it. If it can't be read it will produce a fairly clear error and TBH I anticipate this never happening because it would be a pretty odd situation (either the kernel is badly broken and can't produce the contents of /proc/version or someone has changed the permissions on that file to make it unreadable, which, why?).

[DilumAluthge]

Makes sense to me!

[imciner2]

So, the problem is the string microsoft can appear in non-WSL Linux kernels from Microsoft: microsoft/WSL#423 (comment), since apparently some of their cloud images will also contain that string. Instead, the way that seems to be recommended moving forward is to check for the string WSL inside /proc/sys/kernel/osrelease. This seems to be the current "official" way of detecting it for WSL2, since that change comes direct from the WSL repo and devs.

[mgautam98]

@imciner2 systemd says its official way of detecting WSL https://github.com/systemd/systemd/pull/15325/files#diff-0e08ed2dbc97131dbf8deb9eb7af4163R465-R471 but for me (Inside WSL2) not havingWSL string

$ cat /proc/sys/kernel/osrelease
4.19.104-microsoft-standard

$ cat /proc/version
Linux version 4.19.104-microsoft-standard (oe-user@oe-host) (gcc version 8.2.0 (GCC)) #1 SMP Wed Feb 19 06:37:35 UTC 2020

/proc/sys/kernel/osrelease is returning a substring of /proc/version

I think both should work

[imciner2]

but for me (Inside WSL2) not havingWSL string

I guess that change hasn't trickled down yet. It looks like it was fairly recently done in the upstream WSL code, so it might not have made it into updaes yet. I would suggest adding a comment in the code referencing the ambiguity we get with the microsoft string (that it could be both a cloud server or WSL), and reference that WSL issue. That way if someone reports incorrect detection of something, this is documented and doesn't have to be researched everytime.

[StefanKarpinski]

Won't this test fail if you run it under WSL?

[mgautam98]

I don't know how we are going to test it. Yes these tests will fail inside WSL how are we going to test it?

[StefanKarpinski]

I would test that it's not true when the argument is not :Linux since that's always correct and then add this test for the zero-argument version:

if !Sys.islinux()
    @test !Sys.detectwsl()
end

[mgautam98]

Should the Sys.detectwsl(:Windows) test be removed?

[fonsp]

We could call it Sys.iswsl() instead of Sys.detectwsl()? Then it fits in the list of OS checks, and "detect WSL" sounds like an action, not a question - the name does not suggest that it returns a Bool.

[imciner2]

We could call it Sys.iswsl() instead of Sys.detectwsl()? Then it fits in the list of OS checks, and "detect WSL" sounds like an action, not a question - the name does not suggest that it returns a Bool.

See the discussion that occurred in the linked issue about the naming (#36354). The detect prefix was chosen because this is a runtime detection of WSL, not a compile-time constant like all the other sys.is* functions.

[fonsp]

Ah sorry, I didn't see it - I can see why Sys.is* should be reserved.

But I still feel like detectwsl is semantically off - it sounds like it has return type "Union{WSL, Missing}", not Bool. To me, it is comparable to findnext, getfield and getkey, which all return the found index/object, not whether or not it was found.

Some final suggestions:

• Sys.runtimeiswsl() (and Sys.runtimeiscontainer(), etc)

or we return a struct:

struct RuntimeProperties
	WSL::Bool
	container::Bool
	etc
end

Sys.runtimeproperties()::RuntimeProperties = ...

Just wanted to give some alternatives, but feel free to ignore :)

[StefanKarpinski]

Another possible name would be Sys.checkwsl() if you think that would be clearer?

[fonsp]

Yeah somehow that sounds a lot better... I like it!

[fonsp]

Should it take os::Symbol as an argument? Or is it there for convenience?

[StefanKarpinski]

It's a little weird to pass the symbol in this case since you can't really simulate being on another system with this.

[fingolfin]

How about this? (The 1.11 would have to be changed if this were to be backported to 1.10)

Suggested change

	Runtime Predicate for testing if Julia is running inside WSL.
	Runtime predicate for testing if Julia is running inside
	Windows Subsystem for Linux (WSL).
	!!! note
	Unlike `Sys.iswindows`, `Sys.islinux` etc., this is a runtime tests, and thus
	cannot meaningfully be used in `@static if ` constructs.
	!!! compat "Julia 1.11"
	This function requires at least Julia 1.11.

[fingolfin]

This has sat here for a long time unfortunately. So I don't know if @mgautam98 still has any interest in this? If so I'd be happy to help get it over the line. If not perhaps someone would like to "adopt" it. Otherwise we'll have to close it :-(