Your path may differ.

Update: If you are having this problem, you also need to fix $interpreter in ‘bin/vendors’.

My new web app uses the HTML5 History API (as abstracted by History.js) to smooth out navigation. This proved to be a bit of a challenge because, despite all the Javascript, my app is supposed to act just like a normal web page. The main content of my app is dynamically loaded in an ordinary static-positioned DIV. Thus, the user scrolls the window like normal. Other elements are fixed to the viewport.

One major problem I encountered was that browsers have serious issues with remembering the window scroll position across pages Normally, when a user clicks the back button, they are taken to the previous page at the same place that they left it. Most browsers don’t actually reload the page, but store it in memory, so this happens almost instantaneously. When using the History API, this does not always work. In this post, I detail the problems and my solution.

In Webkit, when the user navigates (using links or the back/forward buttons) the scroll does not change from page to page unless the page is too short to support the scroll value. If the user scrolls on page 1 and clicks to page 2, window.scrollY on page2 is the same as it was on page 1. If they scroll on page 2 and navigate back to page 1, scrollY is now the new value. If page 3 is not long enough to scroll, then scrollY is zeroed. Going back to page 1, it remains 0. I like this behaviour because it is predictable. If I am overriding the browser’s navigation, I’d like to have complete control.

Firefox, however, is somewhat erratic in this area. Generally speaking, it behaves like a browser on a normal website. If scrollY on page 1 is 50 and the user clicks to page 2, scrollY is 0. If they scroll a bit so scrollY on page 2 is now 200, then navigate back to page 1, scrollY will be 50. However, if they navigate to page 3 (which doesn’t scroll) and then back to page 1, scrollY will be 0.

I worked around this by storing the scroll value myself, and scrolling to that on each History statechange. Originally, I only stored the current scroll value while handling the statechange event, but Firefox applies its remembered value before firing the event. This placed me in the awkward position of needing to know the value before the user did anything. I use the window.onscroll event to detect when the window has been scrolled by the user so that it does not store a value in the awkward time where new content is loading. Following John Resig’s example, I store the value every half second using setInterval.

Source code

var loc = {
	currentState: 0,
	states: [
		{scroll: 0}
	],
	hasScrolled: true,
 
	initHistory: function() {
		var History = baseWin.global.History;
		if (History.enabled) {
			// Before binding events, replace the state with the proper ID.
			// Otherwise, strange things happen.
			History.replaceState({state:0});
			// Bind the state change event (Dojo won't do this for some reason)
			History.Adapter.bind(baseWin.global, "statechange", loc.handleStateChange);
			// Bind the anchor click
			on(baseWin.body(), "a[href]:click", loc.anchorClick);
			// Bind the scroll event
			on(baseWin.global, 'scroll', function() { loc.hasScrolled = true; })
			// Store the scroll position every 500ms
			setInterval(function() {
				if (loc.hasScrolled) {
					loc.states[loc.currentState].scroll = baseWin.global.scrollY;
					loc.hasScrolled = false;
				}
			}, 500)
		}
	},
 
	anchorClick: function(e) {
		var History = baseWin.global.History,
			rootUrl = History.getRootUrl(),
			node    = e.target;
 
		// Remove orphaned states
		loc.states = loc.states.slice(0, loc.currentState + 1);
		// Initialize the new state
		loc.states[loc.currentState + 1] = {scroll: 0};
 
		// If there is a span, b, i, etc inside the link, that will be the target node.
		// This loop finds the actual anchor.
		while (node.nodeName.toLowerCase() != 'a') {
			node = node.parentNode;
		}
 
		var url 			= domAttr.get(node,"href"),
			isInternalLink 	= url.substring(0,rootUrl.length) === rootUrl || url.indexOf(":") === -1;
		if (isInternalLink) {
			event.stop(e);
			url = url.replace(rootUrl,"")
			History.pushState({state:loc.currentState + 1},"",url);
		}
	},
 
	handleStateChange: function(e) {
		var History = baseWin.global.History
			state = History.getState();
 
		// Load the state id
		loc.currentState = state.data.state;
 
		// IMPLEMENTATION SPECIFIC XHR STUFF
		// onSuccess: function() {
			baseWin.global.scrollTo(0, loc.states[loc.currentState].scroll);
		// }
	}
}

Naturally, the above code would have to be modified for another application, but I hope that the method is clear and is of use to someone.

Changing CSS visibility breaks things

Several of the problems I had seem to have been caused by various problems with Safari’s handling of visibility. For example, a number of elements in the app use the following CSS transition to fade in and out:

Transitioning Opacity

.hideable {
	opacity: 0;
	visibility: hidden;
	transition: visibility 0.5s linear 0.5s, opacity 0.5s ease;
	-moz-transition: visibility 0.5s linear 0.5s, opacity 0.5s ease;
	-webkit-transition: visibility 0.5s linear 0.5s, opacity 0.5s ease;
	-o-transition: visibility 0.5s linear 0.5s, opacity 0.5s ease;
}
 
.visible {
	opacity: 1;
	visibility: visible;
	transition-delay:0s;
	-moz-transition-delay:0s;
	-webkit-transition-delay:0s;
	-o-transition-delay:0s;
}

I use visibility for two reasons: (1) elements can be sized according to their contents before being shown, (2) visibility can be transitioned. The second reason is important, because otherwise it would require additional Javascript to delay the hiding of the element until after the opacity transition runs. This works in Desktop Safari, Chrome, Firefox, the Android Browser, and Mobile Chrome. In Mobile Safari, however, the element does not get shown until you click on another link.

[Edit: Apparently I was wrong about the Android Browser. I suspect this may be a problem in an older Webkit, since it definitely works in Mobile Chrome.]

I originally solved this by disabling the visibility transition in iOS, but I ran into another problem. Merely toggling visibility breaks Mobile Safari’s new native “overflow: auto/scroll” functionality. I’ll talk about this problem in more detail later. For now, the solution is to disable the transition and use display instead of visibility in Mobile Safari:

Mobile Safari Fix

body.iOS .hideable {
	display: none;
	opacity: 1;
	-webkit-transition: none;
	visibility: visible;
}
 
body.iOS .visible {
	display: block;
}

Overflow scroll breaks on visibility toggle

iOS5 finally introduced scrolling within an element. Previously, this had to be achieved with Javascript (e.g. iScroll). Frankly, I was pretty happy with my Javascript solution, but some variation of the position fixed bug mentioned below was breaking it. The touchable portion of the inner DIV scrolled with the document, rather than remaining fixed.; if the document was scrolled partway down, the fixed element no longer scrolled.

iOS’s solution is fairly simple, just add the following to your CSS: “-webkit-overflow-scrolling: touch;” I imagine it’s done this way because they’re afraid of breaking apps that relied on overflow-scroll just not working. It works pretty well, though it does not use the custom scrollbars. It also does not work if the scrolling DIV is contained in an element that was previously visibility-hidden. This time, it has nothing to do with transitions. The only fix I can find is to use display instead of visibility.

 Javascript scrolling breaks touch events on position-fixed objects

This was a weird one. I put buttons on the right edge of the screen to allow a user to page up and down the document. They were position-fixed (supported in iOS 5!) and they worked. Once. You couldn’t click them a second time. If you clicked one, all the other position-fixed buttons stopped working. If you then scrolled normally, everything started working again. Interestingly, if I just scrolled down 100px, the button would stop working, but if I then touched a spot 100px above the button, it would register. It’s as if Safari has a map of elements for the purpose of registering touch events that stays fixed to the page (ie: position absolute), while the visible representation of the elements stays fixed to the screen. Thus, when the page scrolls, the map goes with it. For whatever reason, Javascript doesn’t trigger a recalculation of where the elements are.

I disabled the buttons in iOS. Apple is aware of this, so a fix may be coming.

It shouldn’t surprise me since it’s a sneaky SEO strategy (all SEO strategies are sneaky) that makes perfect sense. The Doodle ensures that millions of people will run a particular search that day (Mies van der Rohe architecture). If you create an article that uses those terms, then you can drive some extra traffic to your publication. People who come to find out who Ludwig Mies van der Rohe is will stay to read a couple other articles. Still, it saddens me to see news reporting subordinated to Google Doodles. It’s not the architect or the Doodle that’s the news, the news is SEO.

As I get more used to Symfony, I find myself wishing that I had given RoR a longer try back in the day. It’s too late for me to start my current project from scratch again, so PHP it is and will be. But gosh, Ruby looks fun.

This script will repair symlinks after you’ve moved a folder. Change BASEDIR to the current location of your files and OLDBASE to where they used to be. For example, if you moved all your files from /mnt/Backup2 to /mnt/Backup2, then BASEDIR = ‘/mnt/Backup2′ and OLDBASE = ‘/mnt/Backup1′. If you want to test (to make sure it will do what you expect) then change DEBUG to True.

This script will only fix symlinks that point to files/directories within the BASEDIR.

Source code

#!/usr/bin/env python
 
import os
 
# Configuration
 
BASEDIR = '/mnt/NewStorage'
OLDBASE = '/mnt/OldStorage'
DEBUG = False # I recommend a test run first
 
def relink(path):
    old_target = os.path.realpath(path)
    new_target = old_target.replace(OLDBASE,BASEDIR,1)
    if DEBUG:
        print "Relink: " + path + "\n\tfrom " + old_target + "\n\tto " + new_target
    else:
        os.remove(path)
        os.symlink(new_target,path)
 
for root, dirs, files in os.walk(BASEDIR):
    for filename in files:
        fullpath = os.path.join(root,filename)
        if os.path.islink(fullpath):
            relink(fullpath)
    for dirname in dirs:
        fullpath = os.path.join(root,dirname)
        if os.path.islink(fullpath):
            relink(fullpath)

If you save the script as ~/link-fix.py then run it with python ~/link-fix.py.

Ubuntu uses an event-based system called udev that, among other things, can run a script when a USB device is plugged in or unplugged. Local (ie: user-created) udev rules are stored in /etc/udev/rules.d/. Before you can create a rule, you need to know a little about the device that is going to trigger it. Open a terminal and type:

udevadm monitor --udev --environment

Connect your USB device. It doesn’t have to be a USB key, it could be a phone, for example. A bunch of text will come up. Look for “ID_SERIAL” and “ID_VENDOR_ID”. Write down the values of those. You can use other variables if they suit your device better, just change the rules file accordingly.

Create a file called “/etc/udev/rules.d/85-screen-lock-toggle.rules”. Put the following in it, replacing VALUE with the proper value:

File: /etc/udev/rules.d/85-screen-lock-toggle.rules

ACTION=="remove", ENV{ID_SERIAL}=="VALUE", ENV{ID_VENDOR_ID}=="VALUE", RUN+="/usr/local/bin/gnome-lock-enable"
ACTION=="add", ENV{ID_SERIAL}=="VALUE", ENV{ID_VENDOR_ID}=="VALUE", RUN+="/usr/local/bin/gnome-lock-disable"

Next create the scripts to actually disable and enable locking:

File: /usr/local/bin/gnome-lock-enable

#!/bin/bash
 
user=`ps aux | grep gnome-screensaver | head -n 1 | awk '{print $1}'`
 
if [ -n $user ]; then
	GNOME_SCREENSAVER_PROC=`ps xa | grep gnome-screensaver | head -n 1 | awk '{print $1}'`
	export `grep -z DBUS_SESSION_BUS_ADDRESS /proc/$GNOME_SCREENSAVER_PROC/environ`
	su $user -c "gconftool-2 --set "/apps/gnome-screensaver/lock_enabled" --type bool 1"
fi

File: /usr/local/bin/gnome-lock-disable

#!/bin/bash
 
user=`ps aux | grep gnome-screensaver | head -n 1 | awk '{print $1}'`
 
if [ -n $user ]; then
	GNOME_SCREENSAVER_PROC=`ps xa | grep gnome-screensaver | head -n 1 | awk '{print $1}'`
	export `grep -z DBUS_SESSION_BUS_ADDRESS /proc/$GNOME_SCREENSAVER_PROC/environ`
	su $user -c "gconftool-2 --set "/apps/gnome-screensaver/lock_enabled" --type bool 0"
fi

Make the scripts executable:

sudo chmod a+x /usr/local/bin/gnome-lock-enable /usr/local/bin/gnome-lock-disable

Restart udev:

sudo restart udev

Enjoy the convenience.

Here’s a step-by-step way of moving a panel to another screen:

1. Right-click the panel you wish to move and select “Properties”.
2. Uncheck the “Expand” option under the “General” tab.
3. Grab one of the edges of the panel by clicking on the left or right end (top or bottom end for vertical panels).
4. Drag the bar to the desired screen and position.
5. Check the “Expand” option in the “Panel Properties” window and click “Close”.

From Chris Jean’s blog.