proposal: net/http: add constant for content type

[tomocy]

Is it a good idea to add constants for content type of HTTP body like http.MethodPost in HTTP methods?

It might be

const (
    ContentTypeFormURLEncoded = "application/x-www-form-urlencoded"
)

This changes will help us not typo, and get support from editors.

[rsc]

"ContentTypeForm" would be a shorter, better name. Should there be a few of the most common ones here?

const (
    ContentTypeBinary = "application/octet-stream"
    ContentTypeForm = "application/x-www-form-urlencoded"
    ContentTypeJSON = "application/json"
    ContentTypeHTML = "text/html; charset=utf-8"
    ContentTypeText = "text/plain; charset=utf-8"
)

It would be unfortunate if this leads to more requests to add more, like MethodPost etc have.
It would also be unfortunate if people start writing r.FormValue("Content-Type") == ContentTypeText, since that kind of comparison still needs to parse the string.

Thoughts?

[cespare]

It would also be unfortunate if people start writing r.FormValue("Content-Type") == ContentTypeText, since that kind of comparison still needs to parse the string.

This is already an all-too-common mistake and I think that adding the constants would, if anything, make it a little worse.

Before, the issue with code like

r.Header.Get("Content-Type") == "text/plain"

was that the left side needed to be parsed to get the media type value out. With

r.Header.Get("Content-Type") == http.ContentTypeText

both sides need to be parsed. Or worse, with

r.Header.Get("Content-Type") == http.ContentTypeJSON

it happens to work now, but depends on ContentTypeJSON being a media type value without optional parameters.

Because these constants can only be effectively used for setting the Content-Type and not retrieving it, I don't think we should add them. If we do add API for common Content-Types, I think it ought to be opaque types that can be used for both setting and retrieving.

@bradfitz Have you thought about Content-Type usability? I don't see any references to Content-Type in the discussion around http(client) v2.

[bradfitz]

I'd like to see a new (string-backed) type for media types. Then we could do:

r.SetContentType(http.TypeForm)

And because http.ContentTypeForm would be a typed constant, you couldn't write r.Header.Get("Content-Type") == http.ContentTypeJSON and get bitten by mismatched parameters.

A nice place for this type might be the mime package, but https://golang.org/pkg/mime/#ParseMediaType already exists and just returns a string, which is unfortunate.

I'd rather not duplicate it with a newer version in net/http.

[rsc]

@bradfitz suggested something quite different from the original proposal, and no one has replied. It sounds like we should decline this proposal as too little impact / not clear enough solution.

Will leave this open for a week for any final comments; if not, will decline.

[cespare]

I think @bradfitz's proposal would be an improvement over the status quo, and I would use it in a bunch of places.

However, I'd still prefer an API which helps with both common content-type operations: setting a response's content-type and checking a request's content-type. The ideas in this issue only address the former.

That should probably be a different proposal, though, if someone comes up with a concrete suggestion.

[rsc]

Marked this last week as likely decline w/ call for last comments (#31572 (comment)).
Declining now.

[ghostsquad]

Sorry to post on a dead proposal, but I'd like understand more about why this is potentially an issue:

r.Header.Get("Content-Type") == http.ContentTypeText

@cespare mentioned that the right hand side has to be parsed. How is a const string different from a string literal in this regard?

[cespare]

@ghostsquad it's not different in that regard. Whether the RHS is using a constant or a string literal, that code is a little broken.

[ghostsquad]

what's broken about it?

[bradfitz]

what's broken about it?

"text/plain" != "text/plain; charset=utf-8"