cmd/go/internel/get: VCS path regexp omits characters accepted by CheckImportPath

[stub42]

What version of Go are you using (go version)?

$ go version
go version go1.12.2 linux/amd64

Does this issue reproduce with the latest release?

Yes

What operating system and processor architecture are you using (go env)?

go env Output

$ go env
GOARCH="amd64"
GOBIN=""
GOCACHE="/home/stub/.cache/go-build"
GOEXE=""
GOFLAGS=""
GOHOSTARCH="amd64"
GOHOSTOS="linux"
GOOS="linux"
GOPATH="/home/stub/go"
GOPROXY=""
GORACE=""
GOROOT="/snap/go/3540"
GOTMPDIR=""
GOTOOLDIR="/snap/go/3540/pkg/tool/linux_amd64"
GCCGO="gccgo"
CC="gcc"
CXX="g++"
CGO_ENABLED="1"
GOMOD=""
CGO_CFLAGS="-g -O2"
CGO_CPPFLAGS=""
CGO_CXXFLAGS="-g -O2"
CGO_FFLAGS="-g -O2"
CGO_LDFLAGS="-g -O2"
PKG_CONFIG="pkg-config"
GOGCCFLAGS="-fPIC -m64 -pthread -fmessage-length=0 -fdebug-prefix-map=/tmp/go-build922868437=/tmp/go-build -gno-record-gcc-switches"

What did you do?

Attempted to fetch the git repo containing a go package, publicly available at:

• git://git.launchpad.net/~stub/+git/go-eggs
• git+ssh://git.launchpad.net/~stub/+git/go-eggs
• https://git.launchpad.net/~stub/+git/go-eggs

Per go help importpath, the following should attempt a git download using https://, falling back to git+ssh://

go get -v git.launchpad.net/~stub/+git/go-eggs.git

Attempting to force ssh per git help environment does not help

GIT_ALLOW_PROTOCOL=ssh go get -v git.launchpad.net/~stub/+git/go-eggs.git

What did you expect to see?

go get attempt to clone the git repo at https://git.launchpad.net/~stub/+git/go-eggs, and if that failed, fall back to cloning the git repo at git+ssh://git.launchpad.net/~stub/+git/go-eggs.

Per go help importpath, because the import path has a version control qualifier (.git) then no attempt should be made to attempt to get the import over https: to look for a tag.

What did you see instead?

Go get attempts to parse meta tags from https://, which fails. No attempt is made to clone the git repo.

$ go get -v git.launchpad.net/~stub/+git/go-eggs.git
Fetching https://git.launchpad.net/~stub/+git/go-eggs.git?go-get=1
Parsing meta tags from https://git.launchpad.net/~stub/+git/go-eggs.git?go-get=1 (status code 200)
package git.launchpad.net/~stub/+git/go-eggs.git: unrecognized import path "git.launchpad.net/~stub/+git/go-eggs.git" (parse https://git.launchpad.net/~stub/+git/go-eggs.git?go-get=1: no go-import meta tags ())

[bcmills]

Duplicate of #26134

[bcmills]

The regular expression that looks for the .git suffix is here:

go/src/cmd/go/internal/get/vcs.go

Line 1028 in ccd9d9d

re: `^(?P<root>(?P<repo>([a-z0-9.\-]+\.)+[a-z0-9.\-]+(:[0-9]+)?(/~?[A-Za-z0-9_.\-]+)+?)\.(?P<vcs>bzr|fossil|git|hg|svn))(/~?[A-Za-z0-9_.\-]+)*$`,

Testing against that (playground) reveals that the problem here is that the + character fails to match the regular expression, which is more restrictive about + and ~ than our validation for import paths in general:

go/src/cmd/go/internal/get/path.go

Lines 125 to 136 in 3cb92fc

	// pathOK reports whether r can appear in an import path element.
	//
	// NOTE: This function DIVERGES from module mode pathOK by accepting Unicode letters.
	func pathOK(r rune) bool {
	if r < utf8.RuneSelf {
	return r == '+' || r == '-' || r == '.' || r == '_' || r == '~' ||
	'0' <= r && r <= '9' ||
	'A' <= r && r <= 'Z' ||
	'a' <= r && r <= 'z'
	}
	return unicode.IsLetter(r)
	}

[bcmills]

This needs a fix, but it's not obvious to me whether the VCS path should be made more permissive, or CheckImportPath should be made more restrictive. I'll follow up with @rsc and @jayconrod.