Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
internal/core/adt: introduce structure sharing
Structure sharing allows the same node to be used in
multiple positions if the evaluation is known to be the
same. This is particularly the case if a field has a single
reference to another field:
a: b // reference to b can be used as is
b: 2
Structure sharing has the potential to greatly reduce the
time complexity of evaluation in certain conditions. See,
for instance, benchmarks/share.txtar, where a simplified
version of the Billion Laughs attack vector is evaluated
in O(n) time, where n is the number of input nodes.
In practice, there are many cases in CUE where repeated
computation can be avoided, and structure sharing has
been seen as one of the big wins to improve performance.
Note that although structure sharing is purely meant as
a performance improvement, it also alters computation
order and it may cut evaluation short before tricky cycle
situations are encountered. So it may fix some bugs as
a side effect.
Note that structure sharing may also interact with other
performance improvements, resulting in compounding
reductions.
Note that although this change also largely makes
structure sharing work with the API, it is mostly
intended to make it work with the core evaluator only.
We will still need to add thorough tests and
evaluation of the API w.r.t. structure sharing.
This is generally true for the new evaluator, so we
plan to do this towards the end of its development.
Issue #2884
Issue #2854
Signed-off-by: Marcel van Lohuizen <[email protected]>
Change-Id: Ic1c888e9567242d6779093fcb391d6e001881f60- Loading branch information
