TempAlloc Overhaul

[Andrei Alexandrescu]

On 6/11/11 11:26 AM, dsimcha wrote:
> I've overhauled my TempAlloc proposal based on some of the suggestions
> I've received.

Review:

* Mention no safety at disadvantages: deallocating memory may leave 
pointers dangling.

* The synopsis should use scope(exit) TempAlloc.frameFree(); as that is 
a robust idiom.

* newArray -> "For performance reasons, the returned array is not 
initialized." Then call it newUninitializedArray. Make safety the 
default and lack thereof the verbose alternative.

* newArray -> "It is not scanned for pointers by the garbage collector 
unless GC.addRange is called." I'd agree that malloc() shouldn't do so 
because it doesn't know the type, but with newArray you do know the type 
so you could automatically call GC.addRange appropriately. If the client 
doesn't want that, they can obtain it with extra syntax. Since now you 
have two things (uninitialized and noscan) you may want to make them 
function parameters instead of part of the function name.

* I'm a bit unsure about free(). People already have the pointer, so 
free() should use it as a cheap way to check that they're freeing the 
right thing. There should be free(void*), freeLastAllocated(), and 
possibly destroyAndFree(T)(T *).

* segmentSlack() -> segmentAvailable().

* stackCat() is a bit surprising. A better way would be something like a 
TempAlloc appender that accepts a chain().

* In fact looking at tempArray() just below that's all - it can accept 
chain(). No need for stackCat.

* mixin newFrame should be renamed to scopedFrame.


Andrei