os: output CR LF for \n on Windows

[jftuga]

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

$ go version
go version go1.11.1 windows/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
set GOARCH=amd64
set GOBIN=
set GOCACHE=C:\Users\john\AppData\Local\go-build
set GOEXE=.exe
set GOFLAGS=
set GOHOSTARCH=amd64
set GOHOSTOS=windows
set GOOS=windows
set GOPATH=C:\Users\john\go
set GOPROXY=
set GORACE=
set GOROOT=C:\Go
set GOTMPDIR=
set GOTOOLDIR=C:\Go\pkg\tool\windows_amd64
set GCCGO=gccgo
set CC=gcc
set CXX=g++
set CGO_ENABLED=1
set GOMOD=
set CGO_CFLAGS=-g -O2
set CGO_CPPFLAGS=
set CGO_CXXFLAGS=-g -O2
set CGO_FFLAGS=-g -O2
set CGO_LDFLAGS=-g -O2
set PKG_CONFIG=pkg-config
set GOGCCFLAGS=-m64 -mthreads -fno-caret-diagnostics -Qunused-arguments -fmessage-length=0 -fdebug-prefix-map=C:\Users\john\AppData\Local\Temp\go-build222202374=/tmp/go-build -gno-record-gcc-switches

What did you do?

package main
import "fmt"
func main() {
   fmt.Println("test")
}

What did you expect to see?

C:\>test.exe | od -tx1
0000000 74 65 73 74 0d 0a
(notice the Carriage-Return character = 0x0d)

According to https://en.wikipedia.org/wiki/Newline#Representation

A newline under Windows is represented by CR LF, which is 0x0d 0x0a

Please consider adding 0x0d to fmt.Println, Printf, etc.

What did you see instead?

C:\>test.exe | od -tx1
0000000 74 65 73 74 0a
(Carriage Return is missing, but Linefeed is there

[ianlancetaylor]

People seem to have managed so far without this change on Windows. Changing this now would almost certainly break people.

[jftuga]

Can this be changed for Go 2.0?

The longer you wait to make this change, the worse it will be.

[davecheney]

The longer you wait to make this change, the worse it will be.

It's been nine years. I don't think the world is going to end.

[as]

And what will "\r\n" do?

[jftuga]

It will be useful when redirecting a command line program's output to a file.

For example:

C:> dir > files.txt
C:\> myGoPgm.exe > output.txt

The dir command as well as any other command-line program that has ever run under Windows will use \r\n for new lines. However, myGoPgm that uses fmt.Println will only output \n for newlines. Since the expected behavior for Windows command-line programs is \r\n, it would be nice if Go programs did the same thing under Windows as well, but still use just \n for Linux and MacOS.

See also: https://en.wikipedia.org/wiki/Newline#Representation

[ianlancetaylor]

The problem is that sometimes people really do want just \n. It would be a disaster to always turn \n into \r\n; that would break all code that writes a binary file. So it would be necessary to do something like what C does on Windows, and add a O_BINARY flag to pass to os.OpenFile. Since this means that code has to change regardless, much better to require people who care to add an io.Writer that does the transformation for them, such as https://godoc.org/github.com/andybalholm/crlf.

[mattn]

This is my indivisually opinion. I have ported many UNIX applications to Windows so far. In many cases, the tests are failed due to the difference in the line feed code, people had frustrations after ported to Windows too. After I met Go where the line feed code is "\n", I feel this problem seems to became fewer, and I noticed the portability of Go is very nice. Using Go, we were able to port UNIX applications to Windows with just small changes. Recently, Notepad on Windows10 support to open a file that have UNIX line code. I feel that the time will come for Windows to abandon "\r\n" in same as the Mac abandoned "\r".

https://blogs.msdn.microsoft.com/commandline/2018/05/08/extended-eol-in-notepad/

I use bufio.Scanner when reading the output of a command that not known the line feed code.

[as]

@jftuga I am not asking you what \r\n is. Rather, I am asking you what you expect fmt.Fprintln(fd, "\r\n\r\n\n\r\r\n") to do on a Windows box.

Side note: Windows programs write \r\n, but most of them do not break when given \n on the consumer side. Having used windows cmd.exe for longer than I can tolerate, my personal list of programs that break on \n is very short: the interpreter of batch files, for example, is such a program.

Notepad, as mentioned above, was one application that did not work. However, wordpad (write.exe) does. Native command line utilities work too.

[jftuga]

Ideally, Fprintf("\n") would output \r\n on Windows with the exception being if the \n is immediately preceded by a \r then the complete output would be \r\n and not \r\r\n

But I am really not sure about this. Originally, I was only thinking about Println and Printf and not Fprintln

[ghost]

https://nurmi-labs.blogspot.com/2016/11/git.html

Git-2.10.0-32-bit.exe

includes dos2unix.exe and unix2dos.exe

http://sourceforge.net/projects/mingw/files/MSYS/Base/msys-core/msys-1.0.11/

MSYS-1.0.11.exe

contains the awk scripts d2u and u2d

https://nurmi-labs.blogspot.com/2015/10/awk.html

link to UNIX awk

https://nurmi-labs.blogspot.com/2017/05/9pvfs.html

links to pre-compiled watcom-1.9 and -2.0 awk.exe versions

whether the d2u or u2d (above) scripts

work using the UNIX and/or the watcom-1.9 and -2.0 awk.exe versions

I am uninformed on

[jftuga]

Can we simply consider the case of fmt.Println for now? If you are using fmt.Fprintf then you have explicit control. Would it be possible for just fmt.Println to use CRLF under Windows?

[ghost]

I am used to dealing with BOM, CRLF, executable scripts, and file permissions, after having used different Windows subsystems, win32, Interix, and WSL,

[ianlancetaylor]

@jftuga Your suggestion is unfortunately not as simple as it sounds. fmt.Println(...) is exactly the same as fmt.Fprintln(os.Stdout, ...). os.Stdout is a global variable that programs can and do change. So we can't attribute special behavior to fmt.Println; we can only attribute special behavior to the default value of os.Stdout. That leads us straight back to O_BINARY.

If we adopted #13473 (for Go 2) then it would be at least feasible to wrap the default values of os.Stdout and os.Stderr with an io.Writer that turned \n into \r\n. I do not know whether that would be a good idea in the long run.

[ghost]

From Emacs' GUI frontend one can change the end-of-line style on an open file via the second button from the left on the modeline.

Vim is now 27 years old (according to Bram Moolenaar).

Help poor children in Uganda!

I presume the GUI frontend, gvim.exe, likewise as the Linux version, includes in the menu's Edit dropdown, an entry for File Settings, wherein one navigates to File Format, which opens a popup from where one can 'Select format for writing the file', i.e., Unix, Dos, and Mac, respectively LF, CRLF, and CR.

https://nurmi-labs.blogspot.com/2015/10/regexp.html

9pm051031.zip

BOMs if present will show up on the first line of flat files running 'cat filename' using 9term.exe, and can be edited out using sam.exe, using the backspace key.

See also: https://github.com/nurmi-labs/9pm

[jftuga]

@ianlancetaylor

If we adopted #13473 (for Go 2) then it would be at least feasible to wrap the default values of os.Stdout and os.Stderr with an io.Writer that turned \n into \r\n. I do not know whether that would be a good idea in the long run.

I like this idea for Go 2. What is the best way to determine if this is a good long-term idea or not?

[as]

What programs (other than notepad and the batch interpreter) have trouble with pure line feeds on the consumer side? I feel like this is a step back for Go on Windows. It would break many Go ports of unix tools that have deterministic output across operating systems.

[ianlancetaylor]

To be honest, I think it's unlikely that we would make this change in Go 2. The argument in favor would have to demonstrate workflows that do not work today but would start working if we made the change. It's not enough to show that Go programs do not work like Windows programs; we need to know what is actually broken by that fact.

[mvdan]

Here's an argument against modifying the behavior of os.Stdout to handle newlines; what if a program supports writing binary data to os.Stdout? I can think of a handful of programs that can print raw data to a file or to standard output. I imagine those would silently break, if the binary data happens to contain a newline.

[ghost]

https://github.com/golang/protobuf
Go support for Google's protocol buffers

Would the above be relevant in regards to the previous comment?

[as]

encoding/gob moreso

[davecheney]

I vote to close this ticket. There are too many unknowns. For example. If I compile a program on windows and runs on windows should it produce windows line endings for all io.Writers or only just os.Stdout? What about if I compile a program from Linux to goos=windows, what rules apply then? What if I compile a program on Linux and pipe it’s output to a file then move that file to a windows machine? There are just too many edge cases to make a blanket rule like “println uses \r\n on windows” practical. Additionally the fact that we’ve survived nine years without needing to make this change suggests that this is not a place to spend the limited Go 2.0 innovation tokens.

…

On 17 Nov 2018, at 13:06, as ***@***.***> wrote: encoding/gob moreso — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or mute the thread.

[alexbrainman]

If anyone cares, I support @ianlancetaylor

People seem to have managed so far without this change on Windows. Changing this now would almost certainly break people.

and @mvdan argument

what if a program supports writing binary data to os.Stdout

We need to see the problem with what happens now, before we consider changing it.

We should not just follow what https://en.wikipedia.org/wiki/Newline#Representation says.

I also looked to see what gcc does to

printf("a\n");

(it outputs "a\r\n")

and to

printf("a\r\n");

(it outputs "a\r\r\n").

Still that does not sway me towards doing the same in Go.

Alex

[ghost]

Another argument against changing the code might be maintaining consistency across Windows environment subsystems, win32/win64 and WSL.

While the cygnus port (a posix layer above win32/win64) isn't a supported platform for golang, cygwin occasionally shows up in golang issues and/or threads, however I would point out that, IIRC, I.L.T. wrote in summer of 2012 that "...cygwin binaries are considerably slower than Windows binaries".

Google Code Archive
https://code.google.com/archive/p/golang-on-cygwin/source/default/source

I'd also like to see this issue closed.

[creker]

From my experience working with systems with different line endings, any solution that tries to be clever and produce different line endings depending on some logic, creates more problems that it solves. Every one of them has different logic, every piece of code that accepts line endings have different logic. In the end you almost always end up with some glue code that ties incompatible systems together.

I vote we leave Go as is. Being opinionated is good in cases like that. At least we have consistent ecosystem that works the same everywhere. If some ancient piece of code breaks because of that, it's better to fix it or write some glue code. More reasons to write your programs properly from the beginning.

[kardianos]

As a long time Go, Windows, and Linux user, I would view this as disastrous. I do not like programs that try to be smart about line endings. This is especially disastrous when you try to have some type of reproducible output.

This type of logic inevitably must try to distinguish between "text" and "binary" modes. I remember old FTP programs where if you downloaded a program or image in text mode it would corrupt it due to this feature.

I believe this issue should be closed as won't fix.

[robaho]

You may want to review how Java has handled this from its inception. Just define a platform constant LineSeperator then you can create TextReader/Writer types that perform bidirectional encoding as needed. They developer can wrap the standard os.Stdout as needed. Seems the only easy way to not break backward compatibility.

[ghost]

Consider exploring on a Wiki page using PowerShell to output alternate end-of-line style plus BOM and minus BOM, for the win32/win64 environment subsystem, if that really isn't satisfactory, then revisit the current issue.

[jftuga]

OP here. I just want to thank everyone for considering this change. I know it would/could be a big change. Would it be possible to make this a user-defined setting within the fmt package? The setting could be enabled by a function that changes the output of \n. By default, the newline behavior is the same. However, if you were to call this function, then \n would output \r\n on Windows platforms when one of the fmt.Print functions is invoked. What do you think?

fmt.useNativeNewline(true)

[ghost]

the following comment addressed a question, which has remained unanswered

#28822 (comment)

[jftuga]

@forskning: I will work on a list of programs and let you know my results tomorrow.

[mattn]

I prefer to add implementation of Writer to write CR LF when write LF instead of add fmt.useNativeNewline.

ex: https://github.com/mattn/go-textwriter

[ghost]

I've sent an enquiry to someone from the 'PowerShell Team' (a GitHub organization) if they would be inclined chip in.

[ianlancetaylor]

We are definitely not going to add a setting to the fmt package. Sorry.

[jftuga]

I created a text file that contained just LF and compared it to a file that had CRLF endings.

I opened the LF file. MS Word 2016 & Excel read the file OK, but when it saved the file, both programs converted LF to CRLF. I don't see a problem with this - just an observation. VS Code, Notepad++, Sublime Text, Visual Studio 2017 had no problems opening/saving files (as expected). My version of the Windows 10 Notepad did not work properly, but as previously mentioned the very latest version of the Windows 10 fixes this.

Batch files, using cmd.exe, definitely have a problem with just LF endings. I would see this error when reading the the LF file, but not the CRLF file: The syntax of the command is incorrect.

I did some limited testing with PowerShell and it seemed OK with just LF, but I don't know enough about PS to make a definitive statement one way or the other. I only tested Get-Content and [System.IO.File]::ReadLines()

In summary, as long as you are not using cmd.exe, LF seems to be OK an there may or may not be issues with PowerShell. In my case, I do use cmd and noticed this problem.

[ghost]

What about mintty.exe?

[ghost]

runemacs.exe

M-x eshell

outputs CRLF

% cat eshell_out | od -tx1
0000000 74 65 73 74 0d 0a
0000006
%

[mattn]

I want to know about the original problem. Why output of LF is not good for you? And how many cases that is higher priority that should be CRLF than LF? IMHO, in many cases, CRLF that is needed is fewer than LF. At least for Go language, fmt.Println must work as same on all environments since it should be cross compiled with same code. If CRLF is needed, the programmer should use CRLF intentionally, I think.

[ghost]

Looks like the solution is to learn Elisp or PowerShell coding.

I wouldn't mind if this issue stayed open a few weeks to see if any PowerShell users added comments.

[ghost]

#28822 (comment)

If anyone cares, the bug in the 'file' command has been fixed.

file/file@6d90cbf
Avoid over-trimming UCS16 text, and ending up losing the last character.
committed Nov 27, 2018

[jftuga]

This more concisely describes potential issues than what I previously posted:

https://en.wikipedia.org/wiki/Newline#Issues_with_different_newline_formats

[ghost]

@jftuga

here's Haskell's implementation

http://hackage.haskell.org/package/base-4.12.0.0/docs/System-IO.html

[ghost]

Consensus seems against this proposal. A couple of other coding languages' approaches to the subject have been cited in this issue. DOS-style CRLF can be output using existing software provided in Windows 10, so I'm going to suggest closing.

NB: PowerShell Remoting Over SSH

[ianlancetaylor]

We aren't going to do this, so closing.

[jftuga]

OP here, thanks for everyone's consideration on this. I really appreciate it.

[jan4984]

any one, how to send binary to client with ResponseWriter in windows platform? 0a turns to 0d 0a!!

[jan4984]

em, the httpie python client converting it. no problem with golang ResponseWriter.

[mvdan]

This issue is closed. For questions and help, see https://golang.org/wiki/Questions.