Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Use x/sys/windows instead of syscall #197

Open
wants to merge 14 commits into
base: main
Choose a base branch
from

Conversation

TBBle
Copy link
Contributor

@TBBle TBBle commented Feb 11, 2021

This basically happened by accident while I was working on #185. Pretty-much everything used from syscall also appears in x/sys/windows, and a bunch of stuff in the latter isn't in the former, so this is a net deletion of LoC, which is nice.

The only things left that import syscall are the syscall wrappers generated by golang.org/x/sys/windows/mkwinsyscall.

I apologise in advance to reviewers for the second commit. it was one search-and-replace followed by fixing compiler errors, and that turned into "convert the world".

@TBBle TBBle requested a review from a team as a code owner February 11, 2021 12:20
)

//sys bind(s syscall.Handle, name unsafe.Pointer, namelen int32) (err error) [failretval==socketError] = ws2_32.bind
//sys bind(s windows.Handle, name unsafe.Pointer, namelen int32) (err error) [failretval==socketError] = ws2_32.bind
Copy link
Contributor Author

@TBBle TBBle Feb 11, 2021

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I was super-disappointed that windows.Sockaddr is not implementable outside x/sys/windows. On the other hand, it has a buffer that's only 116 bytes, so we couldn't fit a HvsockAddr in there anyway right now (Edit: I can't count, a GUID is 128 bits, not 128 bytes). I guess it's a future problem to add AF_HYPERV and SockAddrHyperV to x/sys/windows directly.

@thaJeztah
Copy link
Contributor

I think #172 is also still pending (moving some changes to upstream golang.org/x/sys)

@TBBle
Copy link
Contributor Author

TBBle commented Feb 11, 2021

Oh, interesting. With a brief look, I suspect there should not be a semantic conflict with #172, because that PR didn't change any of the syscall-sourced types, and I didn't try and replace existing code with x/sys/windows where the types differ, e.g. []byte => *windows.SECURITY_DESCRIPTOR, just the ones where I could literally replace syscall with windows and find the right thing.

I have replaced localFree with windows.LocalFree though, as has #172 in a few places, and I think I've done a pass previously to get rid of the (*[0xffff]byte)(unsafe.Pointer(X))[:Xlen] idiom, since that fails a gc check added in go 1.14. (That latter change might have been in hcsshim, not go-winio though. I've lost track, it was biting me when vendor'd into containerd anyway).

@thaJeztah
Copy link
Contributor

@kevpar @containerplat PTAL

@dcantah
Copy link
Contributor

dcantah commented Mar 10, 2021

All of these look fine (and thank you for doing it) besides the vhd ones as most of the signatures change. When updating the vhd package recently I made an effort to keep using syscall to not pull the rug up from anyone. I think we'd be fine taking this and cutting a release that documents the vhd funcs take in and return the windows equivalents now, but someone chime in with any objections here

@TBBle
Copy link
Contributor Author

TBBle commented Mar 10, 2021

It wouldn't be an issue if we want to keep the vhd package's API using syscall.Handle, it's fairly isolated, so it wasn't caught up in the win32File.handle change that triggered this whole adventure, and it doesn't have any relationships with things in syscall that are differently structured in x/sys/windows.

But the question then becomes "will it ever change" and if so, "when is best to do it?".

I'd lean towards "earlier", while the API is (AFAIK) not otherwise changing, so there's only one upgrade pain-point at a time. And maybe it'll serve as a prompt for the consumers to also migrate from syscall to x/sys/windows. ^_^

@dcantah
Copy link
Contributor

dcantah commented Mar 10, 2021

I think if we cut a release after the change we should be fine. Agreed sooner rather than later, I hope the API for that package is pretty locked in by now.

@thaJeztah
Copy link
Contributor

If we want to "rip off" the bandaid, should we also include #172 in the same release? (perhaps that one needs to be carried as it looks like it needs a rebase)

@dcantah
Copy link
Contributor

dcantah commented Mar 10, 2021

@thaJeztah That sounds ideal. Paging @katiewasnothere

@thaJeztah
Copy link
Contributor

As to "breaking" change; I (hope) the change doesn't go silently; i.e. better to fail hard than "silent"; at least that should bring it to user's attention that something changed.

Technically (assuming "SemVer(ish)"), this project is still "unstable" (pre-1.0), so "anything goes". That said, I would recommend releasing these changes as v0.5.0 (incrementing the "minor" version), and make sure to add a "Breaking change" note in the release notes. Doing so might trigger the curiosity of users ("hey, what's new?"), but more importantly, releasing as v0.5.0 would "reserves" v0.4.x versions for future use.

If things go really bad (some important LTS project using v0.4.x, and there's a critical bug in go-winio), it would still provide the opportunity to do a v0.4.17 release (preventing such project from having to include the breaking change to get the important bugfix). Of course, ideally, such a v0.4.x release would not be needed, but just in case, it probably won't hurt to have it available as an option.

@TBBle
Copy link
Contributor Author

TBBle commented Mar 11, 2021

Yeah, I'm almost certain that at no point has Go let me silently replace syscall.Handle with windows.Handle, despite both being aliases for the same underlying type (i.e. it would be safe if this was silent), so this won't be a silent change for users of the VHD package.

Coming from a C++ background, that's a pleasant surprise, much as it lead to, well, this PR (since I changed one structure, and just kept fixing compile errors...)

@TBBle
Copy link
Contributor Author

TBBle commented Apr 20, 2021

As an update, I was expecting to land this after #172, which I thought was ready-to-go in March, but seems to have stalled. I expect if this lands first, #172 shouldn't need too much reworking, if we want to get this moving.

I also don't mind if this just waits for #172, so we can put them in together (batching the API changes), this one isn't blocking me at this time.

@thaJeztah
Copy link
Contributor

v0.4.18 was tagged (#197 (comment)), so I think it should now be ok to merge the breaking changes (targeting v0.5.x)

wdyt @kevpar @dcantah @katiewasnothere ?

@dcantah
Copy link
Contributor

dcantah commented Apr 20, 2021

The plan was to have one final .4.x tagged release (0.4.17) before this got in and then bump to 0.5.x with this breaking change, but we had one extra requested addition for Moby the other day so we cut another release. This looks fine to merge to me. @katiewasnothere was there anything extra needed for the SecurityAttribute change? Otherwise let's get both of these in and cut a new release.

file.go Show resolved Hide resolved
@TBBle TBBle force-pushed the use-x_sys_windows-instead-of-syscall branch 2 times, most recently from 0ba5488 to 71b2a58 Compare April 21, 2021 11:39
@TBBle
Copy link
Contributor Author

TBBle commented Apr 21, 2021

The plan was to have one final .4.x tagged release (0.4.17) before this got in and then bump to 0.5.x with this breaking change, but we had one extra requested addition for Moby the other day so we cut another release.

I'd like to highlight a versioning issue we hit at moby/moby#42307 (comment), and that we might want to undo an API change from #185, release an 0.4.19, and then redo the API change and merge this and #172.

That's not the only feasible solution to that issue, but I wanted to make sure the discussion was visible, since this PR is "continue/complete the ABI change I introduced in-passing in #185".

@thaJeztah
Copy link
Contributor

I'd like to highlight a versioning issue we hit at moby/moby#42307 (comment)

Yes, I saw the failure, and 🤦 realised that those changes were already in master before the last v0.4 tag was done

katiewasnothere and others added 7 commits April 21, 2021 12:49
This reverts commit bfd5468.

Signed-off-by: Kathryn Baldauf <kabaldau@microsoft.com>
This reverts commit ef753e6.

Signed-off-by: Kathryn Baldauf <kabaldau@microsoft.com>
Signed-off-by: Kathryn Baldauf <kabaldau@microsoft.com>
Add CI github action for testing on push and PR
Signed-off-by: Kathryn Baldauf <kabaldau@microsoft.com>
Even though this repo is normally used as a dependency, there are some
sample binaries and tools built out of here. This just makes sure they still build
in the CI.

Signed-off-by: Daniel Canter <dcanter@microsoft.com>
dcantah and others added 7 commits April 21, 2021 19:01
Build the three binaries in this repo from the ci
…_break

[Temporary] Revert Implement winio.GetFileStandardInfo FileInfo commits
…einfo_break

Revert removal of "Implement winio.GetFileStandardInfo FileInfo" commits
Specifically, this pulls the fix for
golang/go#44538
as we're going to start using this API in place of our own generated
wrapper.

Signed-off-by: Paul "TBBle" Hampson <Paul.Hampson@Pobox.com>
This replaces the use of either syscall or hard-coded magic numbers.

Signed-off-by: Paul "TBBle" Hampson <Paul.Hampson@Pobox.com>
syscall.Handle is core to a _lot_ of structures and APIs, so this
great-big commit contains the entire rest of the conversion from syscall
to x/sys/windows.

I also eliminated a bunch of generated syscall wrappers which had
_direct_ equivalents in x/sys/windows. There's still a bunch which exist
in x/sys/windows and either have worse implementations, e.g.
windows.CreateFile, or somehow expose different APIs, e.g. the calls
used by privileges.go

Signed-off-by: Paul "TBBle" Hampson <Paul.Hampson@Pobox.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

None yet

5 participants