I think I found a bug that started around last Friday (plus or minus one day, I'm in Japan, I live in the future).

I was playing with some simple code, bug following a colored path thanks to a colored marker ( very much like these projects, http://scratch.mit.edu/projects/391731/ and http://scratch.mit.edu/projects/1522852/ ) which stopped working from one day to the next with no apparent reason. (the two projects just mentioned are not working they were supposed to at this time either).

Another project, the most viewed this month, a car parker, ( http://scratch.mit.edu/projects/11704649/ ) also works with a color touching color kind of algorithm and since 5 days ago, dozen of commenters suddenly started complaining that the car crashes for no reason. That leads me to believe I'm not the only one with the issue.

I simplified my code to the max and came up with this

turn right (180) degrees
if <color [#00f] is touching [#f0f]?> then
say [Touching]
else
say [Not touching]
end

With the following setup.

The fushia ball turns around the cross, the image above is what is expected, but when I click on the same code again I got

which is wrong. Note that the speech bubble is still where the fushia ball used to be.

If I click on stop and start from the opposite direction I get

which is correct and if I launch the code again I get

which is wrong.

Was there an small update on Scratch's code or something?
Sorry for being so long, I tried to be detailed.
I'm on Mac OS 10.7.5, I tried Safari and Chrome and saw the same thing.

Hi,

Can you send over the example project? From the code you've provided, I do not see an error. You need to place the “If / Then Else” block in a Forever loop to constantly be updating the “If / Then Else” statement.

forever
if <color [#00f] is touching [#f0f]?> then
say [Touching]
else
say [Not touching]
end

I guess it wasn't very clear. Apologies.

Here is the project: http://scratch.mit.edu/projects/18397922/

Basically for me, when I run it, it keeps saying ‘touching’ even though it's clearly not.

Interesting for me too. I compiled the project using a compiler I ‘ve written and it works as you’d expect. The one area where the performance of the compiled version is lacking is the touching colour and colour touching colour blocks. I've been thinking of ways to improve the performance and this bug is possibly a clue that Scratch uses the optimisation I was considering. To perform the touching function the way the compiled version does it is to render the whole scene but without the sprite/clone which is running the touching script. It then creates a transform to map the sprite to the rendered stage and “walks” through each sprite's pixel and the corresponding rendered stage pixel to see if a sprite pixel is touching a rendered stage pixel of the correct colour. The optimisation I was thinking was that the rendered stage is unlikely to change between touching blocks in the same script and so you don't need to re-render the scene each time. Looks like Scratch is not re-orienting the sprite when I does the test, maybe due to caching that too?

It appears another Scratcher is encountering this issue. I'll submit this bug to the Scratch development team. Thanks again for telling us about this :]