cmd/go: improve error when a package exists in the filesystem but is in a different module

[firelizzard18]

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

$ go version
go version go1.12 darwin/amd64

Does this issue reproduce with the latest release?

Yes (1.12)

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

go env Output

$ go env
GOARCH="amd64"
GOBIN=""
GOCACHE="/Users/me/Library/Caches/go-build"
GOEXE=""
GOFLAGS=""
GOHOSTARCH="amd64"
GOHOSTOS="darwin"
GOOS="darwin"
GOPATH="/Users/me/Source/Go"
GOPROXY=""
GORACE=""
GOROOT="/usr/local/Cellar/go/1.12/libexec"
GOTMPDIR=""
GOTOOLDIR="/usr/local/Cellar/go/1.12/libexec/pkg/tool/darwin_amd64"
GCCGO="gccgo"
CC="clang"
CXX="clang++"
CGO_ENABLED="1"
GOMOD="/private/tmp/test/go.mod"
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 -fno-caret-diagnostics -Qunused-arguments -fmessage-length=0 -fdebug-prefix-map=/var/folders/vy/7b8vfhcx2v7036h99kzs188h0000gn/T/go-build129924831=/tmp/go-build -gno-record-gcc-switches -fno-common"

What did you do?

$ tree
.
├── a
│   ├── go.mod
│   └── somefile.go
├── somefile.go
└── go.mod
$ cat go.mod
module path/should/not/matter
$ cat a/go.mod
module path/should/not/matter/a

go build ./a
(cd ./a && go build)

What did you expect to see?

I expect go build ./a from the project root and go build from the submodule a to behave the same.

What did you see instead?

Adding a/go.mod causes go build ./a and cd ./a; go build to behave differently.

$ go build ./a && echo success
can't load package: package path/should/not/matter/a: unknown import path "path/should/not/matter/a": cannot find module providing package path/should/not/matter/a
$ (cd ./a && go build && echo success)
success

[firelizzard18]

It appears that go build ./sub/path does not build what's on disk, if ./sub/path/go.mod exists, but instead determines what the import path of ./sub/path is and downloads and builds that. This is extremely counterintuitive and problematic.

[nim-nim]

Each module starts at the directory containing its go.mod file, and contains all packages from that directory and its subdirectories recursively, excluding any subtree that contains another go.mod file.

FAQ

So by creating nested modules, you exclude the resulting files from the main module scope.

[mvdan]

Basically what @nim-nim said. Perhaps the docs should be clearer about "nested modules" not being a supported feature. I seem to remember earlier threads about this, but can't find them now. /cc @bcmills @myitcv @thepudds

[bcmills]

We should fix the error message, at least. Pretty sure this is a duplicate.

[bcmills]

#27957 is related, but doesn't mention fixing the error message explicitly.

[bcmills]

CC @jayconrod

[firelizzard18]

I fully understand that "nested modules" are excluded from the "parent module". I have no issue with that. The issues I do have are:

• Running go build on a relative path, when that path is its own module, does not build the code at that path but instead downloads the module and builds that.
• Running go build on a relative path and running go build after changing directory to that path have different results.

So by creating nested modules, you exclude the resulting files from the main module scope.

Sure, fine, that's reasonable and makes sense. But go build ./a seems to be doing this:

1. Look for go.mod at the specified path
2. Read the module name/id from go.mod
3. Instead of building the module at the specified path, download the module to some location (the cache?) and build that.

I don't understand the logic behind that last step. When I run go build <path> on a path that is prefixed with ./, I expect it to build from my local workspace. go is interpreting go build ./a as go build path/should/not/matter/a, which is entirely unexpected and unwelcome.

[gopherbot]

Change https://golang.org/cl/185345 mentions this issue: cmd/go: rationalize errors in internal/load and internal/modload