get_option() and get_blog_option() not as similar as you might expect

[ZappoMan]

So I was just debugging some crazy behavior, and I stumbled into the code to find something I very much did not expect.

I always assumed that:

Code:

switch_to_blog(1);
$val = get_option("foo");
restore_current_blog();

would be return the SAME values as:

Code:

$val = get_blog_option(1,"foo");

And in mu 1.0, it did... but somewhere along the way... things changed...

and now, it turns out that they do not do the same thing...

No, sometime around 1.5 (maybe sooner) the code changed so that when using the cache, these go to totally different cache groups.

the get_option() version goes to the "alloptions" cache, while the get_blog_option() goes to the "site-option" cache (with the blog ID preprended).

This seems like a problem to me... since I can imagine code that assumes that these two calls would return the same semantic object... but alas... you could easily get these things out of sync.

Does anyone else think this is a problem?

Should I report this as a bug in trac?

[drmike]

I would bring it up on the support forums at least and ask for Donncha's input.  Stuff like that, he usually will pop in and answer.

[Andrea_R]

I would totally mention it to get his input.

[ron_r]

The switch to blog version of the function was commented out in 1.3. I didn't check earlier versions.

[ZappoMan]

@ron_r: I think this issue is also related to how the cache is implemented and also whether or not plugins use get_options() or get_blog_options() exclusively.

It's funny because I only found this bug because I had to reimplement some stuff, and in the process of debugging other issues, I noticed this code was written this way. The code itself wasn't actually causing problems for me.

As of now, my code is working fine, because I'm explicitly doing the switch_to_blog() because I've got other things I need to do in the blog context.

But I can just imagine the bugs that could pop up from this.

[ron_r]

It's funny because I only found this bug because I had to reimplement some stuff, and in the process of debugging other issues, I noticed this code was written this way. The code itself wasn't actually causing problems for me.

I could see the potential for issues if update_option was used in contexts where both are being used. update_blog_option updates both. But, I expect update_option is not even aware the other is there.