[czrnb]

> I'd say data races are race conditions, but not all race conditions are data races.

In practice, I think you are almost right. There are some cases which exhibit data races, while being free of race conditions, but I think the linked post overstates the importance of differentiating between data races leading to and free of race conditions.

> That's really multiple operations > [...] > if you deconstruct the actual operations performed by a single line of code as I did here

It is not a given that the increment operation is to be divided into individual operations.

All of this is very dependent on the language (interpreted vs compiled (and how it's compiled)), how the concurrency is implemented (managed by the kernel vs software threads), as well as the ordering model of the hardware (whether the processor takes the liberty of reordering instructions).

The data races you describe lead to a race condition under the assumption that `a` or a derivative is used in some conditional branching later on. Arguably, this is always the case, otherwise, there would be no point in having this variable in the first place. But this is just one example.

Now for an example of a data race without race conditions which, IMO, is a bit more explicit than the one from the linked post. Imagine the following: we want each of our threads to write a timestamp of its execution into a shared variable, in order to know when the last thread was executed. This situation contains a data race by definition, as we do not know which thread will write its value into the shared variable. However, both values are similar enough for our monitoring needs (e.g. correct), and do not lead to race conditions.

[didibus]

Hum... okay. Maybe I'm not understanding what is meant by a data race here.

So your example would be:

  function f() {
    lastTime = Time.now();
  }

And now if we had threads T1 and T2 calling f concurrently, you say this would be a data race?

So I'm confused here. There's no bug, so a data race is not a category of defect the way race conditions are? Both your example and the article's example was a data race that is not a bug. Is there an example of a data race that is not a race condition yet leads to a bug?

[czrnb]

By the article's definition, yes, it would be a data race. With this example, I just wanted to make it more explicit to you what a data race without a race condition could look like.

Truly I'm still not sure why the author had to make a point and write a blog post about the conceptual difference between data races and race conditions.