Could we know if a StructTag key exists, even if it has an empty value?

[Matthew Holt]

I'm thinking of something similar to accessing map keys:

value, ok := someMap[key]

Where `ok` is true or false, depending on if the key exists in the map. This is tremendously helpful.

Similarly, I think it would be really convenient to know if a struct field tag exists, even if its value is "". I'm talking about this function: http://golang.org/pkg/reflect/#StructTag.Get

For instance, to mark a field as required for Martini, you have a field tag like this:

type BlogPost struct {
    Title   string `form:"title" json:"title" required`

Notice "required" is tacked on at the end. This doesn't break the existing implementation of the tag parser, but it would be nice to know if the "required" tag exists, even if its value is empty string "":

tagValue, ok := structField.Get("required");

I realize this is not necessarily backward-compatible, so what about a new function on StructField like .Tag() or .Has() or something?

I'm asking because If this is never going to happen, because otherwise we're going to change the interface to "binding" package in martini-contrib: https://github.com/codegangsta/martini-contrib/issues/60#issuecomment-30576449 to decide upon a permanent tag format that's more verbose and less convenient: 'validate:"required"`.

[Matthew Holt]

Yup, already done. But as one user points out in that issue, it does break Go convention. It is more convenient than `required:"true"` or `validate:"required"` which seems unnecessarily verbose. One of Go's great mantras is brevity, but it's lacking here. I'm willing to submit new code for review, but I'm skeptical about how it would be received.

On Saturday, December 14, 2013 9:25:55 AM UTC-7, Islan Dberry wrote:

A StructTag is a string. You can write your own parser for this string in martini-contrib. 

[Kamil Kisiel]

> One of Go's great mantras is brevity

Citation needed ;)

[Matthew Holt]

Valid point.

On Saturday, December 14, 2013 10:16:05 AM UTC-7, Islan Dberry wrote:

On Saturday, December 14, 2013 8:29:48 AM UTC-8, Matthew Holt wrote:

Yup, already done. But as one user points out in that issue, it does break Go convention. It is more convenient than `required:"true"` or `validate:"required"` which seems unnecessarily verbose.

In the Go convention, the key provides scope for independent users of the struct tag. The tags `required:"true"` and `validate:"required"` do not follow this convention.

The tag should be something like `binding:"required"`.  

[Matthew Holt]

Definitely considered it. We talked at length about how to implement the required indicator. The problem is that "form" is only one way to bind (it could also be JSON which uses the encoding/json package). We want the required indicator to be universal no matter how the request was received.

On Saturday, December 14, 2013 10:28:41 AM UTC-7, Islan Dberry wrote:

On Saturday, December 14, 2013 9:16:05 AM UTC-8, Islan Dberry wrote:

The tag should be something like `binding:"required"`.  

If I understand the code correctly, you are already using the "form" key.  You should do something like:

   Title string `form:"title,required"`   // name=title, required = true

   Foo string `form:",required"`  // name=Foo, required=true

   Bar string `form:"bar"` // name=bar, required=false

[Ian Lance Taylor]

On Sat, Dec 14, 2013 at 8:04 AM, Matthew Holt <matthe...@gmail.com> wrote:
>
> I'm thinking of something similar to accessing map keys:
>
> value, ok := someMap[key]
>
> Where `ok` is true or false, depending on if the key exists in the map. This
> is tremendously helpful.
>
> Similarly, I think it would be really convenient to know if a struct field
> tag exists, even if its value is "". I'm talking about this function:
> http://golang.org/pkg/reflect/#StructTag.Get
>
> For instance, to mark a field as required for Martini, you have a field tag
> like this:
>
> type BlogPost struct {
> Title string `form:"title" json:"title" required`
>
> Notice "required" is tacked on at the end. This doesn't break the existing
> implementation of the tag parser, but it would be nice to know if the
> "required" tag exists, even if its value is empty string "":
>
> tagValue, ok := structField.Get("required");
>
> I realize this is not necessarily backward-compatible, so what about a new
> function on StructField like .Tag() or .Has() or something?

I don't see a persuasive reason for this. StructTag.Get is for the
convention described at http://golang.org/pkg/reflect/#StructTag. You
are describing a slightly different convention. You can write a
decoder for your convention easily enough, but I don't see a strong
reason to change the way the existing one works.

Ian

[Matthew Holt]

Mostly, it would be so that we can use valueless keys in tags where it makes sense without breaking the existing "Get" method.

But the convention of the package name being the key also has merit.

Currently, valueless keys break the "Get" method on StructTag, unless it's at the end. Would it be worth making "Get" more robust to handle the case where a key without a value is found? Right now the behavior is undefined. This would let other authors use struct tags more liberally without changing the standard convention.

For what it's worth, the binding package in martini-contrib has already been altered to conform to the convention.

[Rob Pike]

Instead of changing the standard to match your convention, change your
convention to match the standard.

-rob

[Matthew Holt]

Yup, done.

--
PRIVACY NOTICE: Do not include me in mailings when my private address is visible to all the recipients. Please send the email separately or use BCC instead.

[Naitik Shah]

If you still want it: http://godoc.org/github.com/ParsePlatform/go.structtag

> --
> You received this message because you are subscribed to the Google Groups
> "golang-nuts" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to golang-nuts...@googlegroups.com.
> For more options, visit https://groups.google.com/groups/opt_out.



--
-Naitik