tree a32e282e27dcbb97f422f2bc4cd0d052d91353f0
parent 7110020ebafb4a4cccb7de42234e4d63a27c90d5
author Marcel van Lohuizen <mpvl@golang.org> 1612185439 +0100
committer Marcel van Lohuizen <mpvl@golang.org> 1612269503 +0000

internal/core/adt: more fixes to cycle handling

Note that this also fixes some regressions from v0.2.2:
  cue/testdata/fulleval/046_non-structural_direct_cycles.txtar:
    c1: was a cycle erorr whereas it really was correct
    c2: reported a cycle error, whereas it really was a permanent conflict
This test was ported from v0.2.2 and its v0.2.2. results can be found
in the sections out/legacy-debug and out/def.

Also note that this reverts some of the "fixes" made in Issue #667:
although most reported cycle mis-handling, there were some
other cycles that got inadvertently resolved. Most notably,
to resolve checking the existence of fields, a Vertex needs to
be evaluated to a certain state before a conclusive decision
can be made. If two fields mutually depend on reaching such
a state it cannot be determined. The file
     cue/testdata/cycle/compbottomnofinal.txtar
describes the essence of this. This is arguably a type of
reference cycle, but maybe this should be considered a
different kind of cycle.

Fixes #695

The solution traces references that are used in
nonMonotonic situations (compare against bottom)
and compares the against other evaluation results,
setting traps for conflicts.

This is likely not covering all cases. A true fix to
this issue would require a more recursive evaluation.
This is planned, but will not be done for now.

Change-Id: I0d98cf60eb7757daa9ceda99fcfd01757203b3b7
Reviewed-on: https://cue-review.googlesource.com/c/cue/+/8441
Reviewed-by: CUE cueckoo <cueckoo@gmail.com>
Reviewed-by: Paul Jolly <paul@myitcv.org.uk>
Reviewed-by: Marcel van Lohuizen <mpvl@golang.org>