Daniel Martí

Some updated numbers, as of CUE 655236e96e9f447948b8669263da98c5700c147d: ``` $ time CUE_EXPERIMENT=evalv3=0 cue export ./k8s/amour/list.cue >/dev/null real 0m10.683s user 0m18.254s sys 0m0.456s $ time CUE_EXPERIMENT=evalv3=1 cue export ./k8s/amour/list.cue >/dev/null ^C (stopped...

I have finally been able to reduce a bit of CUE which seems to OOM my machine where evalv2 definitely did not: https://github.com/cue-lang/cue/issues/3509

#3509 is fixed but that doesn't fix your issue. Quickly trying to narrow down its cause, I ran into a related regression, https://github.com/cue-lang/cue/issues/3511.

Still seems to happen as of 0005c227d164576e39f8436b577cfcf9d98924f4.

I again reduced part of this project into another performance issue to be fixed: https://github.com/cue-lang/cue/issues/3633

Here is another reduction, similar to #3633, yet not the same: ``` import "list" items: [...#JSONSchemaProps] #JSONSchemaProps: { props?: [string]: #JSONSchemaProps repeat0?: [...#JSONSchemaProps] repeat1?: [...#JSONSchemaProps] repeat2?: [...#JSONSchemaProps] repeat3?: [...#JSONSchemaProps] repeat4?:...

Significant progress; I am now able to finish evaluating with evalv3, even if it's still quite a bit slower than evalv2: ``` $ for n in 0 1; do time...

I ran this again, as of the latest version of the automata and cue repos, and got the following stats: ``` $ for n in 0 1; do time CUE_EXPERIMENT=evalv3=$n...

I think this issue overlaps with https://github.com/cue-lang/cue/issues/3881 too; the pprof shows me: ``` flat flat% sum% cum cum% 3530ms 25.05% 25.05% 3530ms 25.05% cuelang.org/go/internal/core/adt.transitiveMapping 1930ms 13.70% 38.75% 3610ms 25.62% runtime.scanobject...

With https://review.gerrithub.io/c/cue-lang/cue/+/1213494, we are now looking much better, even if still a bit slower: ``` $ time CUE_EXPERIMENT=evalv3=0 cue vet -c ./k8s/magiclove/list.cue real 0m6.335s user 0m10.852s sys 0m0.247s $ time...