On Sun, Oct 21, 2018 at 8:34 PM, Fei Ding <fding...@gmail.com> wrote: > > I am studying golang source code, literally `go/src/runtime/chan.go`, the > question is in function `makechan`, when it comes to malloc memory for > channel with/without pointers, you guys did it as: > > > case elem.kind&kindNoPointers != 0: > // Elements do not contain pointers. > // Allocate hchan and buf in one call. > c = (*hchan)(mallocgc(hchanSize+uintptr(size)*elem.size, nil, true)) > c.buf = add(unsafe.Pointer(c), hchanSize) > default: > // Elements contain pointers. > c = new(hchan) > c.buf = mallocgc(uintptr(size)*elem.size, elem, true) > > So, my question is why to use two different ways to do this, (continuous vs > separated) ? and what's the difference/relationship between malloc and new?
The hchan struct is designed so that it doesn't contain any pointers that the garbage collector needs to know about, as explained in the comment a few lines above the code you quote. If the channel elements don't contain any pointers, then the garbage collector doesn't have to scan them. So in that case we can allocate a single large block of memory to hold the hchan struct and the channel buffer, and the garbage collector will never to scan that large buffer. That tends to give better memory cache behavior and slightly lessens the load on the garbage collector. When the channel elements do have pointers, we allocate the channel buffer separately, as that is easier than building up a GC structure that handles both the hchan struct and the buffer. The key difference is that in the no-pointer case we pass nil as the type argument to mallocgc, and in the pointer case we pass the element type. Calling new is equivalent to calling mallocgc(sizeof_type, type, true). The only reason to call mallocgc is when the runtime wants to vary one of those arguments. Ian -- You received this message because you are subscribed to the Google Groups "golang-nuts" group. To unsubscribe from this group and stop receiving emails from it, send an email to golang-nuts+unsubscr...@googlegroups.com. For more options, visit https://groups.google.com/d/optout.
