Finalizer support in extent. #307
Conversation
sdlib/d/gc/extent.d
Outdated
| assert(index < slotCount, "index is out of range!"); | ||
| assert(hasFinalizer(index), "No finalizer is set!"); | ||
|
|
||
| return cast(void*) (*last64Bits >> 16 & AddressMask); |
There was a problem hiding this comment.
Ok, it is in the free space because you shift, but it's not if you don't shift. (and, incidentally, if you don't shift, it where it always was).
There was a problem hiding this comment.
Right, the idea here is that now the encoded freespace is to be found below the finalizer (if one exists), and so can occupy 1 byte if fits in 1 byte, or 0 bytes if there is no freespace (as it did before introducing finalizers.)
There was a problem hiding this comment.
@deadalnix AFAIK this piece is complete; can merge? Or do you see any further issues? (e.g. would you prefer that the 2 final bytes get always reserved for the freespace when finalizer is enabled, to trade the space saving for a small speedup?)
sdlib/d/gc/extent.d
Outdated
| cast(ushort*) (address + getUsableSpace(index) - 2)); | ||
| } | ||
|
|
||
| size_t getUsableSpace(uint index) { |
There was a problem hiding this comment.
Not only that shouldn't really be necessary, but this is wrong. This is the second time incorrect code is in there and passes tests. So I'll let add tests for boundary condition and break this down until it is obvious none of these remain. Relying on me to find all the problem is not a solution.
sdlib/d/gc/sizeclass.d
Outdated
| @@ -100,6 +100,22 @@ unittest isAppendableSizeClass { | |||
| } | |||
| } | |||
|
|
|||
| // Determine whether given size class supports finalization. | |||
| bool isFinalizableSizeClass(uint sizeClass) { | |||
| return (getSizeFromClass(sizeClass) & 0x1f) == 0; | |||
There was a problem hiding this comment.
This is probably hopelessly inefficient.
sdlib/d/gc/extent.d
Outdated
| assert(index < slotCount, "index is out of range!"); | ||
| assert(hasFinalizer(index), "No finalizer is set!"); | ||
|
|
||
| return cast(void*) (*last64Bits >> 16); |
sdlib/d/gc/tcache.d
Outdated
| @@ -806,3 +806,38 @@ unittest arraySpill { | |||
| testSpill(16384, [0, 1, 2, 500, 16000, 16383, 16384]); | |||
| testSpill(20480, [0, 1, 2, 500, 20000, 20479, 20480]); | |||
| } | |||
|
|
|||
| unittest finalizers { | |||
There was a problem hiding this comment.
There is no change in tcache, why is there a test in tcache?
|
Superceded by #311 |
No description provided.