Some threading semantics are desirably non-rigorous. What happens when we have to define every aspect of the parallel execution of our program?This article is being composed in a browser based on the Mozilla codebase. It is inherently threaded as it is running under the X windowing system, and the hardware is multicore.
Say that I am browsing a search hit list and I click a link to document a://b.c . If I'm faster (== nondeterminism) than the load of the new document, which can be delayed by network factors, I can open a new tab, or change the destination of the current tab, via rapid middle and left clicks on the original document from which I launched the request for a://b.c .
Is this desirable behavior? Is it an artifact of sloppy threading? A message slipped into the queue is allowed to interact out-of-order with the most plausible semantics of the user interaction. I.e. it seems reasonable to assume a://b.c will load before any links in any document being displayed will be honored and that no clicks links from the document being left behind will be recognized. This is not the case.
One can even poll the developers' intentions without shedding much light on this paradoxical threading, benign though it might be in a browser. It turns out that the developers' intentions are phenomenologically uninteresting compared to the significance of the context in which those intentions were shaped. That context is a state of the art in which the convenience of leaving such interactions undefined heavily outweighed the trouble of defining them.
Indeed the simplicity and loose coupling of behavior somewhat mimic familiar physical systems. It is entirely possible that, in addition to being the only practical project plan given the state of the art, that the effect of the nondeterministic Mozilla link clicking behavior when a document load is already underway is comfortingly normal to a user.
The champions of more conscious multicore largely share a preference for banishing the undefined in favor of the rigorous. Is increasing rigor likely to improve user interactions, or make them increasingly fine-grained, settings-heavy and exasperatingly trivial in its demands on the consciousness of the user?