1052837 - pngcrush and pngquant all the images in our tree

Status: RESOLVED INVALID

(Firefox for Android Graveyard :: Theme and Visual Design, defect)

Description

[Wesley Johnston (:wesj)]

We should be doing this when we check things in, but a quick scan shows we haven't been. Running both on just mobile/android saves about 200K on the APK size. Running them on toolkit gave me another 100K, but getting that patch checked in is probably more work.

I'm going to do one more run with the --speed 1 flag and then I'll upload a patch.

Comment 1

[Wesley Johnston (:wesj)]

Attachment: Patch

I have a feeling there's some reason these aren't done and this is going to be r-, but until then, here's a patch!

For myself in the future, a little bash script to do this:
#!/bin/sh
#modified from http://davidwalsh.name/pngcrush-directory

for png in `find $1 -name "*.png"`;
do
	echo "crushing $png"

	cp "$png" temp.png
	pngquant --speed 1 temp.png
	pngcrush -reduce -brute temp-fs8.png "$png"

	# mv temp.png "$png".old
	rm temp.png
	rm temp-fs8.png
done;

Comment 2

[Richard Newman [:rnewman]]

The reason we haven't done this for every image is that quanting causes banding; when I did a full pass, I only accepted quanted images with fewer than 256 colors, or where there was no visual difference under 8x magnification.

See Bug 959833 Comment 25.

(This was a pretty laborious process.)

Additionally, I *think* the Android build process crushes PNGs automatically. I would be interested to see the size difference without quanting every image -- manually crushing might not be worthwhile.

Probably most of the size saving from this patch comes from the Firefox logo itself (in its various sizes), which responds very poorly to quanting, and which I omitted from Bug 959833.

So: please let's make sure that these changes don't regress anything visually (i.e., only accept lossless quanting), and then verify that we're saving APK size by crushing versus just quanting new images since Bug 959833.

Comment 3

[Richard Newman [:rnewman]]

Comment on attachment 8471886 [details] [diff] [review]
Patch

f+ on principle; see last comment.

Comment 4

[Wesley Johnston (:wesj)]

ckitching found some other space savers. I'm not chasing this anymore, and based on rnewman's comment, I don't think anyone should.