Fix stack overflow from start set calculation

[olsaarik]

This fixes #71808, which was caused by a change I introduced in #71234 making a calculation for the possible starting characters of matches be top-down recursive instead of bottom-up.

The bottom-up computation is re-introduced with one improvement: instead of a separate field for the start set, the set field for Singleton nodes is re-used, since for those the start set coincides with the set. This also allowed making the field non-nullable.

[ghost]

Tagging subscribers to this area: @dotnet/area-system-text-regularexpressions
See info in area-owners.md if you want to be subscribed.

Issue Details

---

This fixes #71808, which was caused by a change I introduced in #71234 making a calculation for the possible starting characters of matches be top-down recursive instead of bottom-up.

The bottom-up computation is re-introduced with one improvement: instead of a separate field for the start set, the set field for Singleton nodes is re-used, since for those the start set coincides with the set. This also allowed making the field non-nullable, which is preferable in the common case where the TSet type is ulong.

Author:	olsaarik
Assignees:	-
Labels:	area-System.Text.RegularExpressions
Milestone:	-

[olsaarik]

Still some semantic bug here.

[stephentoub]

Thanks!

[stephentoub]

Still some semantic bug here.

Meaning it's still stack overflowing? Or some other test is going to fail?

[olsaarik]

Still some semantic bug here.

Meaning it's still stack overflowing? Or some other test is going to fail?

Stack overflow is fixed, but something around anchor handling broke. Debugging.

[olsaarik]

The bug (both old and new) should be fixed now. All tests pass locally for me now.