Fix client-side location not visiting (hard reloading) when returning 409 StatusConflict upon Inertia asset version change.

[the-goodies]

A small refactor to fix client-side not reloading when Inertia has a new asset version. Client side gets a 409 response and does nothing with it - doesn't make a full visit to X-Inertia-Location.
I think a culprit might be that either a response body is not empty or that Content-Type is set to application/json or both of those things.

Now when returning only 2 headers: Vary: X-Inertia and X-Inertia-Location: ... with no other headers and an empty body, it works well and makes a full visit.

Also, I made an early return for this part:

if w2.StatusCode() == http.StatusOK && w2.IsEmpty() {
    i.Back(w, r)
    return
}

[romsar]

Now I think that we have to move version check before next.ServeHTTP(w2, r). I guess that if the version is different, we should not get to this part of the code like in the official adapter https://github.com/inertiajs/inertia-laravel/blob/1.x/src/Middleware.php. What do you think?

[romsar]

Now I think that we have to move version check before next.ServeHTTP(w2, r). I guess that if the version is different, we should not get to this part of the code like in the official adapter https://github.com/inertiajs/inertia-laravel/blob/1.x/src/Middleware.php. What do you think?

No, I was wrong.
https://github.com/inertiajs/inertia-laravel/blob/1.x/src/Middleware.php#L86
Version check should be after all handlers.

[the-goodies]

Yap, version check is after invoking a handler.

I think a culprit might be that either a response body is not empty or that Content-Type is set to application/json or both of those things.

Actually, the real issue was setting X-Inertia: true header (set when calling Inertia.Render).
If we remove it before returning 409, and leave everything else (non-empty body, Content-Type: json, etc,) it still does full reload.

This early return part I made is wrong, because if the request happens to be PUT, PATCH or DELETE, it won't get a correct http.StatusSeeOther response, but instead a default http.StatusFound.

if w2.StatusCode() == http.StatusOK && w2.IsEmpty() {
    i.Back(w, r)
    return
}

So probably we should restore this part (remove early return) and move defer i.copyWrapperResponse(w, w2) before it, like this:

	// Determines what to do when the Inertia asset version has changed.
	// By default, we'll initiate a client-side location visit to force an update.
	//
	// https://inertiajs.com/asset-versioning
	if r.Method == http.MethodGet && inertiaVersionFromRequest(r) != i.version {
		i.Location(w, r, r.URL.RequestURI())
		return
	}

	// Our response writer wrapper does have all needle data! Yuppy!
	//
	// Don't forget to copy all data to the original
	// response writer before end!
	defer i.copyWrapperResponse(w, w2)

	// Determines what to do when an Inertia action returned empty response.
	// By default, we will redirect the user back to where he came from.
	if w2.StatusCode() == http.StatusOK && w2.IsEmpty() {
		i.Back(w2, r)
	}

        // The PUT, PATCH and DELETE requests cannot have the 302 code status.
        // Let's set the status code to the 303 instead.
        //
        // https://inertiajs.com/redirects#303-response-code
	if w2.StatusCode() == http.StatusFound && isSeeOtherRedirectMethod(r.Method) {
		setResponseStatus(w2, http.StatusSeeOther)
	}
})

[romsar]

Thanks! I'll try to find time to write tests for this.

[romsar]

Thanks! I'll try to find time to write tests for this.

On current version hard reloading for me is still not working.
I just made some tests, refactoring and fixes that delete Vary and X-Inertia headers in Location method here: #16

[the-goodies]

Damn, it really didn't work. My first commit worked (which I tested), because it returned before even invoking a handler.
Later commits moved version check after a handler, but I thought since we are not writing back from w2 to w and early returning, we are safe... But since w2 is using the same underlying header map that w is using, X-Inertia: true header being set on w2 automatically is set on w too.

[romsar]

Damn, it really didn't work. My first commit worked (which I tested), because it returned before even invoking a handler. Later commits moved version check after a handler, but I thought since we are not writing back from w2 to w and early returning, we are safe... But since w2 is using the same underlying header map that w is using, X-Inertia: true header being set on w2 automatically is set on w too.

Can you confirm that everything is fine on the latest version now? I've done a couple of tests and everything is working well for me.

[the-goodies]

Yes, works well now (v1.3.4) . Tested my previous versions also (v1.3.2 and v.1.3.3) and they don't work.

[romsar]

Yes, works well now (v1.3.4) . Tested my previous versions also (v1.3.2 and v.1.3.3) and they don't work.

Okay, thanks for reporting the problem.