os: IsNotExist returns true for invalid file names

[perillo]

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

go1.9.2 linux/amd64

Does this issue reproduce with the latest release?

Yes

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

GOARCH="amd64"
GOBIN="/home/manlio/.local/bin"
GOEXE=""
GOHOSTARCH="amd64"
GOHOSTOS="linux"
GOOS="linux"
GOPATH="/home/manlio/.local/lib/go:/home/manlio/code/src/go"
GORACE=""
GOROOT="/usr/lib/go"
GOTOOLDIR="/usr/lib/go/pkg/tool/linux_amd64"
GCCGO="gccgo"
CC="gcc"
GOGCCFLAGS="-fPIC -m64 -pthread -fmessage-length=0 -fdebug-prefix-map=/tmp/go-build134836490=/tmp/go-build -gno-record-gcc-switches"
CXX="g++"
CGO_ENABLED="1"
CGO_CFLAGS="-g -O2"
CGO_CPPFLAGS=""
CGO_CXXFLAGS="-g -O2"
CGO_FFLAGS="-g -O2"
CGO_LDFLAGS="-g -O2"
PKG_CONFIG="pkg-config"

What did you do?

https://play.golang.org/p/7xo02WfeHR

Also, see https://groups.google.com/forum/#!topic/golang-nuts/MZ_u50ZIB_I

What did you expect to see?

file does not exist

What did you see instead?

file exist
stat xxx...xxx: File name too long

[perillo]

This is probably a duplicate of #18974

[ianlancetaylor]

@perillo I don't think it's a dup. #18974 was about ENOTDIR. This is about ENAMETOOLONG. I think it would be reasonable for IsNotExist to return true for ENAMETOOLONG; if we see that error, we know for sure that the file does not exist.

[slrz]

Responding to ENAMETOOLONG probably warrants a different approach than with ENOENT/ErrNotExist. I don't think it should be covered by IsNotExist. I always thought of the latter as "a syscall returned ENOENT(or a Go function returned an error with similar semantics)". ENAMETOOLONG doesn't match that pattern.

The code looks very misleading to me: just because IsNotExist returns false on an error returned from stat, you can't conclude that the file exists. The call might've also hit some wildly different error.

[mvdan]

Shouldn't the issue title read false instead of true?

[ianlancetaylor]

@slrz Upon reflection, I think you're right. One could imagine a deep set of directories, a/b/c/d where let's assume that each letter is actually 255 characters long. Then opening a/b/c/d/e might fail with ENAMETOOLONG, whereas os.Chdir("a/b/c") followed by opening d/e might succeed. At least, that is my current understanding. If that is indeed possible, then I agree that ENAMETOOLONG does not necessarily mean that the file does not exist.

[mattn]

I don't think this should be fixed. non-nil err does not mean file not exists.

https://play.golang.org/p/BnbP5Lik-l

[perillo]

@mattn https://play.golang.org/p/MPqPCmPcjE

The linux man(3) page about errno defines ENAMETOOLONG as "Filename too long"
but the man(2) page about stat defines it as "pathname is too long"

From https://en.wikipedia.org/wiki/Filename and https://en.wikipedia.org/wiki/Path_(computing) it seems that pathname and filename are equivalent, so I don't understand why in my code os.Stat returns ENOENT instead of ENAMETOOLONG.

[tv42]

ENAMETOOLONG means the kernel refuses to deal with your input, and thus provides no information on whether a file exists or not.

@perillo, you need to handle your errors, of which IsNotExist is just one scenario. @mattn's example https://play.golang.org/p/BnbP5Lik-l is the right way to do it (apart from unnecessary nesting).

As far as I can tell, there is no bug here.

[taruti]

Please don't change the functionality of os.IsNotExist for ENAMETOOLONG or the various Windows cases.

[bradfitz]

Sounds like we don't want to change anything here and there's no bug.