Remove outdated fields in PlanConstraints

[qinsoon]

PlanConstraints are constant for each plan. But some of its fields are unused, or superseded by other mechanisms.

moves_objects

mmtk-core/src/plan/plan_constraints.rs

Line 11 in 3dbdd7a

pub moves_objects: bool,

moves_objects is superseded by may_move_objects() in PlanTraceObject, which is a constant value based on the trace kind. It is helpful for plans like Immix, for which, some traces (defrag) move objects and other traces (fast) do not move objects.

gc_header_bits and gc_header_words

mmtk-core/src/plan/plan_constraints.rs

Lines 12 to 13 in 3dbdd7a

	pub gc_header_bits: usize,
	pub gc_header_words: usize,

We now have metadata specs, like mark bit, log bit. We no longer need a plan to specify the GC bits/words they need.

num_specialized_scans

mmtk-core/src/plan/plan_constraints.rs

Line 14 in 3dbdd7a

pub num_specialized_scans: usize,

This is unused. This seems related with object scanning, which is now in the bindings.

max_non_los_copy_bytes

mmtk-core/src/plan/plan_constraints.rs

Line 21 in 3dbdd7a

pub max_non_los_copy_bytes: usize,

This is currently unused. We need to decide whether we need to check object size before copying (currently we assume we won't). The difference is that in one case, we could copy nursery->mature or nursery->LOS, and in the other case, we only copy nursery->mature (large objects are directly allocated into LOS' logical nursery). A related issue: #452.

needs_linear_scan

mmtk-core/src/plan/plan_constraints.rs

Line 30 in 3dbdd7a

pub needs_linear_scan: bool,

Java MMTk uses this in object model. I am not sure how it works. We need to further check this. However, for MMTk core, I assume we will allow linear scan in some policies and when global_alloc_bit is enabled. So we may need to make sure this value is set properly.

needs_concurrent_workers

mmtk-core/src/plan/plan_constraints.rs

Line 31 in 3dbdd7a

pub needs_concurrent_workers: bool,

Currently unused.

generate_gc_trace

mmtk-core/src/plan/plan_constraints.rs

Line 32 in 3dbdd7a

pub generate_gc_trace: bool,

This should be removed. It is related to GCTrace in Java MMTk, and we do not plan to port it.

[k-sareen]

I think, to me at least, part of why these values are useful is if I just want to compare two different plans. Comparing MarkCompact to SemiSpace, I can immediately see that oh MarkCompact uses 1 extra header word and also requires an extra scan in comparison to SemiSpace, instead of having to read its source code.

Though I'll note that we should fix how gc_header_words is used. This is something that I encountered when I was trying to add +1 extra word to header for SemiSpace. Simply increasing the gc_header_words in the plan constraint didn't work since currently the fastpath and everything is hardcoded only for MarkCompact. I will be submitting an issue for this for both mmtk-core and mmtk-openjdk.

[qinsoon]

#1028 removed the following fields:

• gc_header_bits/words
• num_specialized_scans
• generate_gc_trace