Le and Gulwani's paper does cite Little, but the refutation is a bit weak.
> Little and Miller propose a code completion tool that synthesizes the most likely Java expression in a code context from a set of keywords [23]. Smart-Synth is different, in that it synthesizes a complete script and does not require extra contextual information.
From the video, a "complete script" appears to be 1 or 2 simple expressions. And not requiring contextual information sounds like it can't take advantage of contextual information.
You put in a query and get out some code. Concrete code not connected to the query anymore. The code doesn't shift out from under you later, if that's what you're implying.
Why wouldn't it be used in a "proper piece of code"? You ask Visual Studio/XCode/Eclipse for something and it gives you back a snippet that you can use as a starting point.
Pretty neat. I noticed something a bit off about the "tint a random photo blue" example around 2:22. Two of the three search results are fine, but the second search result is
pics := media -> pictures
pic := pics -> random
pic -> tint (colors -> sepia)
c := colors -> blue
All the components are there but not in the right order -- it still tints with the default color.
Interesting research. Of course, you could always hack the Home button to act as a Command key and hold the device in landscape orientation without having to give up a piece of the screen.
This system could be used together with speech recognition (Siri, Cortana, etc. personal agent) to automate tasks and write little apps on a mobile device.
[+] [-] krilnon|11 years ago|reply
Le and Gulwani's paper does cite Little, but the refutation is a bit weak.
> Little and Miller propose a code completion tool that synthesizes the most likely Java expression in a code context from a set of keywords [23]. Smart-Synth is different, in that it synthesizes a complete script and does not require extra contextual information.
From the video, a "complete script" appears to be 1 or 2 simple expressions. And not requiring contextual information sounds like it can't take advantage of contextual information.
[+] [-] opless|11 years ago|reply
[+] [-] chton|11 years ago|reply
Interesting technology, but I doubt it will ever replace or even be used in a proper piece of code.
[+] [-] jameshart|11 years ago|reply
[+] [-] tree_of_item|11 years ago|reply
Why wouldn't it be used in a "proper piece of code"? You ask Visual Studio/XCode/Eclipse for something and it gives you back a snippet that you can use as a starting point.
[+] [-] webwarrior|11 years ago|reply
[+] [-] isbadawi|11 years ago|reply
[+] [-] Leszek|11 years ago|reply
[+] [-] agumonkey|11 years ago|reply
For instance, static touch controls grid are pretty efficient, see:
http://www.malacria.fr/pdf/gutwin14-fasttap.pdf http://www.malacria.fr/video/fasttap-video.mp4
[+] [-] Uncompetative|11 years ago|reply
[+] [-] frik|11 years ago|reply
[+] [-] robgibbons|11 years ago|reply
[+] [-] covi|11 years ago|reply
[+] [-] beemoe|11 years ago|reply
[+] [-] itamarhaber|11 years ago|reply