Bug 29761 – [GTK] performs a POST when refreshing a view that was obtained with a GET

• Component: WebKitGTK
• Status: RESOLVED FIXED

[Xan Lopez]

Quoting from https://bugzilla.gnome.org/show_bug.cgi?id=595348:

I use a web application that handles editing items by having the client POST to
a URL that, when requested, performs the edit. It then redirects the client to
the original URL for viewing the item.

If I perform an edit, and press Refresh after being redirected, epiphany does a
POST to the redirected-to URL rather than a GET.

With the gecko backend, pressing refresh would have resulted in a GET to the
redirected-to URL.

Here's an example.

localhost - - [16/Sep/2009 12:30:17] "GET /view HTTP/1.1" 200 517
 [ now i'll mark the items on the page as read ]
localhost - - [16/Sep/2009 12:30:57] "POST /view/mark_read HTTP/1.1" 302 274
 [ the response redirects the client ]
localhost - - [16/Sep/2009 12:30:58] "GET /view HTTP/1.1" 200 517
 [ fine so far, now I'm back to the view page. I'll now hit refresh ]
localhost - - [16/Sep/2009 12:31:35] "POST /view HTTP/1.1" 404 226
 [ ouch!]

The two problems that are, I guess, related are:

 1. The final request shoudl have been a GET
 2. I was not prompted for confirmation for resumbitting the POSTed information

This happens with epiphany 2.7.92 and webkit 1.1.14. It does not hapen with
epiphany 2.26.3 and xulrunner 1.9.0.14.

[Xan Lopez]

I think this causes automatic posts in bugzilla if you follow a link from a bug after posting, and then come back. Not good :)

[Dan Winship]

I suspect ResourceHandleSoup::restartedCallback needs to do:

    request.setHTTPMethod(String(uri->method));

along with the existing request.setURL, before calling willSendRequest(). (SoupSession will have changed the SoupMessage's method when restarting it in this case.)

[Xan Lopez]

Created attachment 40454 [details]
redirectreload.diff

Patch with Dan's suggestion plus a test for this.

[Gustavo Noronha (kov)]

Comment on attachment 40454 [details]
redirectreload.diff

Thanks!

[Xan Lopez]

The newly added test breaks other test, as the extra line in success200.html adds a new empty line in ~20 test results. Need to either change those or do this in some other way.

[Xan Lopez]

I moved the triggering HTML to a new file, so the old tests won't break. Landed as r49077.