No description, website, or topics provided.
Switch branches/tags
Nothing to show
Clone or download
d1manson Merge pull request #16 from JasonALorance/patch-1
Update app.html to allow authorization
Latest commit 3399bdd Aug 20, 2018

Drive Notepad

A webapp for editing text files in Google Drive, implemented entirely on the client-side.

If you want general information and help using Drive Notepad, please go to the main website.

The application is available live on github.

This document is only meant as a development reference.

General Notes

If, for some reason, you decide to fork this and try and make it work you will need to register various things in the Google Developer Dashboard. Other than that, the only thing you need is node, you can then build with the command npm run build.

See the Google api-explorer for details of RESTful APIs.

Icons are from, then cropped with this online tool, and base64 encoded for css with this other online tool.

I use python -m SimpleHTTPServer for serving content, but you can do as you wish.

In the Google Develeoper Console you need create OAuth2 2.0 client credentials and enable some APIs - Drive API, Drive SDK, and Google Picker API.

All the interesting cloud-based stuff goes through the google javascript client library, which lives in the global gapi. In most cases there are two ways to use a particular API: you can either load a specific API's javascript methods and make use of them; or you can use the generic request method, which helps you "manually" constructing the RESTful request. We use the manual approach for reading and writing files, but the load approach for realtime stuff and picker, and sharer dialogs.

Occasionally chrome mistakenly caches source maps. If that happens, the easiest thing to do is go into the minified js file and add a query string to the source map url (which is specified in a special comment at the end). Normally you won't need to do this, but it's annoying when it does happen.


When the page is loaded we either (try to) create a new document, or (try to) load an existing file. Only one file can ever be loaded, if you need to load/create a new file you have to refresh the page. This makes life a little simpler.

Everything is attached to the dn global object apart from a few basic utility functions in utils.js. The widget's panes are manifested as module/closure things and are held as dn.file_pane, dn.settings_pane, dn.find_pane, etc. They each have their own js file, and they all have .on_document_ready methods which are called from the primary document_ready function in app.js. Each module is responsible for displaying the given pane's current settings and making it interactive, but on the whole they do not make changes outside of the pane (the find tool is an exception as it changes the editor text, scroll position and markers). In general we try and use some form of MVC paradigm for all panes - see section below. app.js contains an important miscellany of functions that are not-specific to individual widget panes.


There are (at least) two separate MVC systems at play in the app, one for the file's metadata and one for the application's setttings. Interested parties can subscribe to a particular model's changes using .addEventListener.

The main model is dn.g_settings. At some point after the page loads, this becomes a Google Realtime API AppDataDocument Model, before that it is a simple home-made object that behaves very similarly. When the realtime model becomes availble, we call .transfer_to_true_model on the mock version, which issues any required change events and then registers the old event listeners with the new model. The home-made object uses localStorage for non-personal settings values so that it can restore them immediately on page load. Also, when migrating to the real model, some settings give the cloud model prioroty and some give the local model priority, this prevents excessive jumpiness for highly-visible things that don't matter that much, ie. the widget's position.

A second model is dn.the_file, which is an instance of dn.FileModel (defined in file_model.js). This is a relatively simple object which backs the metadata shown in the file pane. The only complexity is that when you change some of the values it has to recompute some of the data, e.g. if you change the file's title it has to recompute the message about the detected syntax and, unless a syntax was explicity specified by the user, it must update the chosen syntax. For simplicity, saving (to the server) is handled by the controllers in dn.file_pane, it's just easier to tie the saving action to user actions rather than arbitrary updates on the model.

app.js is, among other things, a view for almost all of dn.g_settings, and controller for some of it (e.g. widget position), it is also a view for most of dn.the_file (e.g. the document title and syntax choice). dn.settings_pane is a view and controller for a subset of settings. dn.file_pane is a view and controller for dn.the_file. dn.help_pane is a view and controller for just dn.g_settings['help_inner'].

dn.find_pane is a bit more complicated: all the settings roughly obey an MVC pattern, backed by dn.g_settings, but the values in the input boxes and currently selected result are implemented in a more ad-hoc manner.

Making async API requests

I have tried to wrap most requests in ES6 Promises, or variations on that. An important promise-like object is dn.pr_auth, this is a SpecialPromise instance, which can be resolved and rejected multiple times. As with regular promises, whenever it is resolved the callbacks registered with .then(success) are triggered, with each being triggered exactly once, even if dn.pr_auth is resolved multiple times. And, as with regular promises, if dn.pr_auth has already been resolved at the moment that the .then(success) is registered, it will be triggered immediately. Finally, the dn.pr_auth has two special event listeners: on_error and on_success, these are called every time the authentication process fails or is successful (respectively).

This behaviour allows us to have actions that wait on the auth token being valid, but if an auth error occurs at a later date we can set the promise to invalid and begin the re-authentication process. For simplicity, we lump network errors (i.e. no internet) together with auth errors, that is dn.pr_auth is resolved when we have internet and authentication and is rejected when either is found to be unavailble. Any actions added after the failure point will await the resolution of the new promise. [The implementation could possibly have been written in terms of ES6 Promises rather than raw JS, but the JS implementation is fairly simple so has its merrits.]

As alluded to in the previous paragraphs, these API calls can fail due to sudden invalidation of the auth token. When that happens we want to re-authenticate and then re-run any pending API calls. This is quite a complex pattern, so it was wrapped into a utilty function, named until_success, and then posted as a SO answer to a related question. Here is an example of how it is used in this app:

until_success(function(succ, fail){
           .then(succ, fail);
    console.log('succeeded getting user info.')
    console.log("failed to load user info")

In this code, dn.show_user_info is called once the API request on the line above succeeds, which in turn is only called when the auth token is available. If the auth token becomes available but is then mysteriously invalidated, the API request will be made and will fail. This will cause that "iteration" of the until_success Promise to fail, invoking dn.filter_api_errors. This handler will decide whether it is worth re-attempting the request, and if some form of re-authentication is neccessary it will start that process. If the handler decides the error is "fatal", then the until_success "loop" exits with whatever the error was, and the .catch handler has to deal with it. If and when the loop finally suceeds, the .then handler is called.

Note how if the dn.filter_api_errors function decides it is worth reattempting the request, the next iteration of until_sucess will start "immediately", but will hang at resolve(dn.pr_auth) until authentication is achieved. It could be that the user needs to manually log in, or it could be the token has become invalid somehow and needs to be refreshed, or it might be that some kind of non-auth transient network error has occured. To guard against the possibility of issuing excessive numbers of automatic reauth requests there is an roughly exponential backoff process, which in the limit will only issue a request once per minute. Note however that this backoff is specific to the automatic reauthentication, not to the until_success machienery.

Another important Pomise-like thing is dn.pr_file_loaded, this simply gets resolved soon after the page loads, when either an existing file is loaded or a new file is created. If neither of those things ever succeed then this is never settled. The fact that this is a promise, allows us to put it in the save chain, and the user can then actually issue save requests before the file is loaded, which is vaugely helpful when creating new files.


The saving system allows for the user to issue multiple requets in quick succession, so that multiple requests are pending at the same time. As the requests return, we check the server's version number to check that they were resolved in the correct order. If anything ended up out of order then we issue a correction save until the order is as we wanted it to be. The things that can be saved are the file body, and the various pieces of metadata shown in the file pane. A single request can contain one or more of these elements. We track the server's version number for each element separately, so we can make minimal corrections as needed.

Note that if there are other users/devices saving at the same time as the current user then there is no guarantee that they will see the same order of changes. The only guarantee is that if user X makes change A and then change B, then once user X receives confirmation of saving being completed, we know the server will defintiely not hold the value of change A. It may hold the value from some other user or it may hold the change B. Also note that rather than simply holding value A then value B, it may have held value B, then A, then B', with B' identical to B but issued in a later "correction" request.


It took a while to sort out focus/blur behaviour. In the end it is now relatively neat and hopefully simple to get your head around.

dn.g_settings.set('pane', 'pane_find') and dn.g_settings.set('pane_open', true) do not mess with the focus themselves. Any code that uses that must explicitly decide whether or not it wants to focus on the input, calling dn.find_pane.focus_on_input(), if it whishes to.

Calling dn.g_settings.set('find_goto', bool), also does not mess with the focus, it just renders the inactive version of goto/search.

We basically have the same setup for goto and for search, with goto having a less meaty implementation, so it's easier to start by looking at that.

When the input gets the focus, the goto/search operation is performed, when the input is blurred it sets the info text to "inactive". If the blur events is moving the focus to null, then at the end of the blur event we redirect the focus to the editor.

dn.find_pane has special functions for producing exactly the right focus behaviour for find/replace/gtoto keyboard shorcuts, and these functions are registered in keyboard.js. Special behaviour when pressing Esc, but when the focus was on the editor, is implemented in keyboard.js.

The rules for cycling and modifying the find history list are a bit fiddly - see the update_search_history function for how we choose to make modification. The main thing to note is that this function is only called on keydown (with ctrl-up/down) events in the find input, and when using the ctrl-f shortcut to focus on find with a new search string.

Ace editor

Currently there are a few additions/changes to ace, which have been implemented as monkey patches in files called js\patch_ace*.js. At some point it might be better to write them up properly in a fork of ace. It's easy to build ace, so that shouldn't be a problem, but modifying the code "well" is a bit tricky because there's not a lot of documentation about its internals.