I'm currently working on a rather interesting... project. I have a client who wants to allow form uploads (from a page presented on their server) specifically to their own google drive account. The platform being used is essentially LAMP.
Single (pre-authenticated) google drive account. Multiple otherwise anonymous upload sources (users).
They do not want users to be required to have their own google accounts (rules out simply using Picker on the user's own drive files).
They want some degree of backwards browser compatibility, such as IE8 (rules out XHR to form the post using HTML5's file API to read the filedata). They don't want to use flash/etc due to potential compatibility issues with certain mobile browsers.
What is working:
- Authenticating (getting a refresh token, storing, using it to get access tokens as needed)
- Uploading a file to the account without metadata
- Result of file upload being sent to hidden iframe
- Catching the iframe load event via jquery to at least know something has happened
Problems:
- The REST API upload endpoint does not support CORS: there is no way to access the result iframe directly. (see: Authorization of Google Drive using JavaScript)
- The return from a successful upload is only raw JSON, not JSONP.
- There is seemingly no way to host anything with proper headers to open via browser on the googleapis.com domain, so easyXDM and similar multi-iframe with cross origin workaround communication javascript approaches are ruled out.
- There is no way to embed a callback URL in the POST from the submit, the API does not allow for it.
- The Picker displays errors on trying to upload if you pass it an Oauth2 token that is not for a user who is also authenticated in their browser (assumedly via cookie). Strangely enough you can show files from the Oauth2 token's matching account, but other than in a browser instance where the target Oauth2 token's account matches the already logged in user any file uploads fail with an ambiguous "Server rejected" message. This happens with all files and file types, including files working in an authenticated browser instance. I assume it's an authentication flow/scope issue of some sort. I haven't tried diving the Picker source.
- All of the javascript Google Drive API upload examples seem to rely on using HTML 5 to get the file data, so anything of that nature seems to be ruled out. While files are uploaded, there's no way other than guesstimating which file came from which user, since we can't get the file object ID from the result in our inaccessible iframe. At best we could make a very rough time based guess, but this is a terrible idea in case of concurrency issues.
- We can't set the file name or any other identifier for the file (not even a unique folder) because the REST API relies on that metadata being sent via JSON in the post request body, not via form fields. So we end up with file objects in the drive with no names/etc.
- We can't create the file with metadata populated server side (or via jquery/XHR, or the google javascript API client) and then update it with a form based upload because the update API endpoint exclusively works with PUT (tested).
- We can't upload the files to our local server and then send them to google (proxy them) as the php ini is locked down to prevent larger file uploads (and back to restrictions imposed on using HTML5 or flash for why we can't chunk files/etc).
All of this has been both researched and to varying degrees tried.
At the moment this is going on hold (at least it was a useful way to learn the API and gain a sense of its limitations) and I'm just going to implement something similar on dropbox, but if anyone has any useful input it would be lovely!
e.g. is there any way to get this working with Drive? Have I overlooked something?
I also realize that this is probably essentially a less than intended use-case, so I'm not expecting miracles. I realize that the ideal flow would be to simply allow users to upload if necessary to their own google drives and then have them grant file access to our web app (via Picker or API+our own UI), but this becomes a problem when not all of our own users are necessarily already google account users. I know that google would OBVIOUSLY prefer we get even more people to sign up with them in order to have this happen, but making people sign up for a google account to use our app was ruled out (not out of any prejudice on our part, it was just too many added steps and potential user hurdles). Even simply having them sign in to google if they did have accounts was deemed unwanted for the basic LCD feature functionality, although it's likely to be added as an additional option on top of whatever becomes the base solution.