go/printer: confusing indentation for nested structs

[tbertelsen]

Please answer these questions before submitting your issue. Thanks!

1. What version of Go are you using (go version)?
   go version go1.6.1 linux/amd64
2. What operating system and processor architecture are you using (go env)?
   GOARCH="amd64"
   GOBIN=""
   GOEXE=""
   GOHOSTARCH="amd64"
   GOHOSTOS="linux"
   GOOS="linux"
   GOPATH=""
   GORACE=""
   GOROOT="/usr/lib/google-golang"
   GOTOOLDIR="/usr/lib/google-golang/pkg/tool/linux_amd64"
   GO15VENDOREXPERIMENT="1"
   CC="gcc"
   GOGCCFLAGS="-fPIC -m64 -pthread -fmessage-length=0 -gno-record-gcc-switches -fdebug-prefix-map=/tmp/go-build255917305=/tmp/go-build"
   CXX="g++"
   CGO_ENABLED="1"
3. What did you do?
   Run gofmt on a file containing this code, that has confusing indention

    tests := []struct {
        name  string
        input *StructA
        want  string
    }{
        {"empty", &StructA{
            NestedB: &StructB{},
        },                       //  <- Confusion indention
            "",                  //  <- Confusion indention
        },
    }

1. What did you expect to see?
   The indention was changed to something less confusing.
2. What did you see instead?
   The indention is what gofmt intends, even though it is confusion.

[mvdan]

This is expected. It won't be as confusing if you don't cram {"empty", &StructA{ in a single line.

[tbertelsen]

It might be expected, but I don't find it really readable. Don't you agree, that the code is relatively easy to misunderstand?

It can very well be that it is not the intention that is the source problem, but the line breaks. I will then suggest that gofmt fixes the linebreaks, for example something like:

        {
            "empty", &StructA{
                NestedB: &StructB{},
            },
            "",
        },

or

        { "empty",
            &StructA{
                NestedB: &StructB{},
            },
            "",
        },

[tbertelsen]

FWIW: Andrew Gerrand (@nf) finds the current behavior "surprising".

[mvdan]

Perhaps forcing a newline in this edge case is an option. I'll poke around gofmt to see if I can find anything.

[josharian]

FWIW, I also found this behavior surprising. I expected the line after "NestedB" to be indented. I can't think of any other cases offhand in which gofmt has such mismatched indentation.

cc @griesemer

[tbertelsen]

@mvdan Thanks for taking a shot at this :)

[mvdan]

I've got a working solution with some tests, will post a CL shortly. This is my first tinkering with go/printer so comments would be appreciated.

[ianlancetaylor]

CC @griesemer

[gopherbot]

CL https://golang.org/cl/24882 mentions this issue.

[nigeltao]

To play devil's advocate, I don't find the existing behavior surprising. Let's label some lines from 0 to 4:

0:  {"empty", &StructA{
1:      NestedB: &StructB{},
2:  },                       //  <- Confusion indention
3:      "",                  //  <- Confusion indention
4:  }

Note that there are two open-curlies on line 0. Let's call them 0a and 0b.

The close-curly on line 2 is vertically aligned with line 0. They both have the same indentation. Note that line 0 is the one containing the matching open-curly, namely 0b. This seems consistent to me, similar to how the matching close-curly in:

func foo() {
    etc
}

is vertically aligned with the start of the line containing the matching open-curly. Similarly for struct literals in general.

As for the "" on line 3, again that seems consistent to me. That is a field of the struct literal (the struct type is an anonymous struct), and the general rule is that fields are indented more than the literal itself, i.e. its defining curlies: 0a and the one on line 4.

I think that gofmt is at least being self-consistent here, and applying simple rules. As for the ugliness of the gofmt'ed output, personally, I'd format it differently:

tests := []struct {
    name  string
    input *StructA
    want  string
}{{
    "empty",
    &StructA{
        NestedB: &StructB{},
    },
    "",
}, {
    etc
}}

Everything seems consistent to me: every close-curly that starts a line is vertically aligned with the line containing the matching open-curly. Every field of a struct literal is at equal indentation, and a greater indentation than its surrounding struct literal.

[mvdan]

I think that gofmt is at least being self-consistent here, and applying simple rules.

I agree that it's being consistent - this is why I initially said "this is expected, you can get a less confusing output by changing the input".

Everything seems consistent to me: every close-curly that starts a line is vertically aligned with the line containing the matching open-curly. Every field of a struct literal is at equal indentation, and a greater indentation than its surrounding struct literal.

If you look at my CL, I have done something very similar. In particular, @tbertelsen's first suggestion:

        {
            "empty", &StructA{
                NestedB: &StructB{},
            },
            "",
        },

I think this is slightly better as it is less intrusive. I would add the minimum number of newlines to avoid the indent confusion. Forcing every field to its own line seems unnecessary to me, especially as go/printer would not do that otherwise.

By the way, this issue should probably be retitled to concern go/printer and not gofmt.

[griesemer]

Fixing this issue correctly is non-trivial. For details see my comments on the CL above. The necessary changes require "measuring" the layout of composite literal elements (go/printer/nodes.go contains a nodeSize method for that purpose) and also collecting layout information (e.g. does an element itself contain indentation). With this information we can then decide whether an extra newline is warranted before such an element, if it occurs on the same line as the opening '{' of the containing composite literal. The extra newline will then force correct indentation for the rest.

The nodeSize method will have to be expanded while maintaining it's current functionality. The results need to be cached ("memoized") so we don't end up in exponential behavior with deep nested composite literals.

I suspect the resulting changes may be too large, now that we're in the freeze. I'll give it a shot on Monday to see how bad it is, but I'm going to mark this 1.8 Maybe for now.

[gopherbot]

CL https://golang.org/cl/33294 mentions this issue.

[griesemer]

@mvdan Here's a first step: https://go-review.googlesource.com/#/c/33294/. This CL adds the code to compute whether a composite literal element contains indentation. One should be able to use this for a complete fix. I haven't quite figured out a clean and correct approach for the rest of the code. It also needs some cleanup. Feel free to give it a shot.

[mvdan]

Thanks @griesemer, that helps.

Given that this would mean redoing the CL from scratch, I suggest we mark this for 1.9. I still want to give it a shot, I just don't think it's a good idea to rush it or get such a non-trivial fix in at this stage of the freeze.

I'll be sure to get another CL in before the 1.9 tree opens in a few months.

While at it, do you have any thoughts regarding other indented lists, such as call arguments? For example: https://play.golang.org/p/LMAfTqtN2O

I think we should be consistent with this "avoiding confusing indentation" fix. At the same time, it seems like the two fixes would be in different parts of the code and should come in separate CLs.

[griesemer]

@mvdan I agree that argument lists should behave the same if we do this.

[griesemer]

Too late for 1.9.