[MLton-commit] [MLton/mlton] 47a781: Fix bug in handling of weak objects during mark-co...

['Matthew F. v. MLton-c. <mlt...@ml...>]

  Branch: refs/heads/master
  Home:   https://github.com/MLton/mlton
  Commit: 47a781c3cd8c242c98466786c695e3ba1f605c00
      https://github.com/MLton/mlton/commit/47a781c3cd8c242c98466786c695e3ba1f605c00
  Author: Matthew Fluet <mat...@gm...>
  Date:   2020-09-20 (Sun, 20 Sep 2020)

  Changed paths:
    M CHANGELOG.adoc
    M runtime/gc/dfs-mark.c

  Log Message:
  -----------
  Fix bug in handling of weak objects during mark-compact GC

During a DFS marking, a weak object should only be linked if its
objptr field is a valid object pointer (and not `BOGUS_OBJPTR`).

This bug was introduced by commit ec2a3afa2, which changed
`is{Objptr,Pointer}` predicates to `is{Objptr,Pointer}InHeap`
predicates.  This was motivated by the introduction of static heaps,
because objects in the static heaps are not traced; in particular,
objects in the immutable static heap are in read-only memory, and
therefore cannot be marked and unmarked.

However, the semantics of these predicates are subtly different.  The
`is{Objptr,Pointer}InHeap` return `TRUE` when the argument is a
non-pointer (i.e., when `is{Objptr,Pointer}` returns `FALSE`).

Commit 2d46cc122 fixed many of the `isPointerInHeap(p)` predicates to
`isPointer(p) and isPointerInHeap(p)`, but missed the `isObjptrInHeap`
that guards the linking of a weak object.


  Commit: e9f6cf4f992eace1357a5e6d84d4c5752ca73b1d
      https://github.com/MLton/mlton/commit/e9f6cf4f992eace1357a5e6d84d4c5752ca73b1d
  Author: Matthew Fluet <Mat...@gm...>
  Date:   2020-09-21 (Mon, 21 Sep 2020)

  Changed paths:
    M CHANGELOG.adoc
    M runtime/gc/dfs-mark.c

  Log Message:
  -----------
  Merge pull request #415 from MatthewFluet/dfs-mark-weak-ptr-fix

Fix bug in handling of weak objects during mark-compact GC


Compare: https://github.com/MLton/mlton/compare/8857346cece4...e9f6cf4f992e