You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Copy file name to clipboardExpand all lines: docs/src/FAQ.md
+10Lines changed: 10 additions & 0 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -97,3 +97,13 @@ The list here is incomplete, but is vetted for quality.
97
97
98
98
-["Evaluating Derivatives"](https://dl.acm.org/doi/book/10.5555/1455489) (by Griewank and Walther) is the best book at least for reverse-mode.
99
99
It also covers forward-mode though (by its own admission) not as well, it never mentioned dual numbers which is an unfortunate lack.
100
+
101
+
## Is removing a thunk a breaking change?
102
+
Removing thunks is not considered a breaking change.
103
+
This is because (in principle) removing them changes the implementation of the values
104
+
returned by an rrule, not the value that they represent.
105
+
This is morally the same as similar issues [discussed in ColPrac](https://github.com/SciML/ColPrac#changes-that-are-not-considered-breaking), such as details of floating point arithmetic changing.
106
+
107
+
On a practical level, it's important that this is the case because thunks a bit of a hack,
108
+
and over time it is hoped that the need for them will reduce, as they increase
109
+
code-complexity and place additional stress on the compiler.
0 commit comments