Struct field, error and non-name xyz on left side of :=

[Lukáš Zapletal]

Hello all,

I know this error has been discussed here many times, but I want to focus on one particular scenario:

type Cache struct {

  file os.File

  // ...

}

cache := new(Cache)

cache.file, err := os.Open(filename) // here the error from %subj%

Now, I would like to know, what is the best approach here - I want Go to determine Error type for me while I still want to initialize both on the same line. What is the best practice here?

Thanks and Happy New Year!

LZ

[minux]

those two goals can't be met at the same time. 

either you give up the first one (the type of err is error, and it's well-known):

var err error

cache := new(Cache)

cache.file, err = os.Open(filename)

or you give up the the 2nd one:

cache := new(Cache)

f, err := os.Open(filename)

cache.file = f

btw, i suggest you write a constructor function for type Cache, for example

func NewFromFile(filename string) *Cache

[jaspe...@kreuzwerker.de]

Hmmm, I've just run into this same problem myself of assigning using ':=' to a struct field. 

Has this really been discussed many times before? I can only find one other thread referring to this issue, which ended in someone saying it might be a good addition to add later: https://groups.google.com/forum/#!searchin/golang-nuts/struct$20field$20non-name/golang-nuts/ItYSNKqt_kA/NbDaqs_4YK0J

Is it possible to generate a little more discussion on why this would not be a good addition? It seems like a nice convenience to me...

[Jan Mercl]

On Thu, Nov 28, 2013 at 3:04 PM, <jaspe...@kreuzwerker.de> wrote:
> Hmmm, I've just run into this same problem myself of assigning using ':=' to
> a struct field.
>
> Has this really been discussed many times before? I can only find one other
> thread referring to this issue, which ended in someone saying it might be a
> good addition to add later:
> https://groups.google.com/forum/#!searchin/golang-nuts/struct$20field$20non-name/golang-nuts/ItYSNKqt_kA/NbDaqs_4YK0J
>
> Is it possible to generate a little more discussion on why this would not be
> a good addition? It seems like a nice convenience to me...

I think the problem is that := is actually not an assignment
operator[0] at all. Thus allowing foo.bar := baz would be IMO
surprising at least.

[0]: http://golang.org/ref/spec#Short_variable_declarations

-j

[jaspe...@kreuzwerker.de]

Are we talking about the same thing Jan? I may not have been very clear in my first post, but I was referring to the example that Lukáš gave originally of:

cache.file, err := os.Open(filename)

where there is at least one new variable (err) and also the struct field which is known.

I agree that something like: foo.bar := baz should not work.

[chris dollin]

On 28 November 2013 16:10, <jaspe...@kreuzwerker.de> wrote:

Are we talking about the same thing Jan? I may not have been very clear in my first post, but I was referring to the example that Lukáš gave originally of:

cache.file, err := os.Open(filename)

where there is at least one new variable (err) and also the struct field which is known.

I agree that something like: foo.bar := baz should not work.

Then cache.file, err := whatever shouldn't work.

The lhs of a short declaration is a list of identifers to be (re)declared.

spinkle.sponkle isn't an identifier. (Whether or not there's a "new variable"

declared is a different matter.)

Chris

--
Chris "allusive" Dollin

[Jasper Timm]

In what definition of 'identifier' does spinkle.sponkle not exist? (http://golang.org/ref/spec#Identifiers) Am I missing something fundamental here?

I think there being a 'new variable' is the whole idea here. In my mind at least, ':=' for multiple entries on the LHS translates to

 ':=' for new terms 

and 

'=' for everything known. 

Is that wrong?

[chris dollin]

Identifiers don't contain dotsvin go.
Chrisatabusstop.

[Jasper Timm]

Fair enough.

But given that, is 'it's not an identifier' enough of a reason? Why is it not ok to extend this to struct fields as well?

[Gustavo Niemeyer]

On Thu, Nov 28, 2013 at 3:06 PM, Jasper Timm <jaspe...@kreuzwerker.de> wrote:
> In what definition of 'identifier' does spinkle.sponkle not exist?
> (http://golang.org/ref/spec#Identifiers) Am I missing something fundamental
> here?

Yeah, it's a selector rather than an identifier:

http://golang.org/ref/spec#Selectors

With that said ...

> I think there being a 'new variable' is the whole idea here. In my mind at
> least, ':=' for multiple entries on the LHS translates to
> ':=' for new terms
> and
> '=' for everything known.
>
> Is that wrong?

It's not how it works today, but it doesn't feel like a totally
unacceptable perspective either. I actually wished that would work a
few times too, and IIRC none of the main developers have demonstrated
strong disagreement to that. It is probably a "let's wait a bit"
situation.

I suggest opening an issue, so it's properly tracked and responded to.


gustavo @ http://niemeyer.net

[Gustavo Niemeyer]

On Thu, Nov 28, 2013 at 3:53 PM, Gustavo Niemeyer <gus...@niemeyer.net> wrote:
> It's not how it works today, but it doesn't feel like a totally
> unacceptable perspective either. I actually wished that would work a
> few times too, and IIRC none of the main developers have demonstrated
> strong disagreement to that. It is probably a "let's wait a bit"
> situation.
>
> I suggest opening an issue, so it's properly tracked and responded to.

I did it: http://golang.org/issue/6842


gustavo @ http://niemeyer.net

[Jasper Timm]

Thanks Gustavo, beat me to it ;)