Hacker News Reader

Comment by brandonbloom | Hacker News Reader

Fun story!

The same idea applies more generally as well with so-called "region-based" memory allocation. A sub-process allocates incrementally from a pool, then when that process ends, the entire pool is freed at once. Optionally, garbage collection or other memory freeing can happen during the process. Additionally, regions can be arranged hierarchically for processes and sub-proesses, recursively.

The easiest way to take advantage of this is, if for example, you have a short-lived command line tool. Maybe you don't even bother to free any memory at all, as the operating system is going to free it all for you when you're done anyway.

View on HN

Game engines often use region-like allocators for data that only lasts a frame -- check out the "tagged heap" in http://twvideo01.ubm-us.net/o1/vault/gdc2015/presentations/G... for example.

The (new in c++17) polymorphic allocators allow this, as you can instantiate a monotonic_memory_resource, and use it to construct all the regular C++ containers with it for the duration of the remaining scope - this works just fine as long as no memory is needed outside of that scope.

...so preoccupied with whether or not they could, they didn’t stop to think if they should.

That's cool. Do the C++ containers now contain a pointer to the monotonic_memory_resource, or is there some magic that lets the local calls to push_back, etc., know about the allocator in that local scope?

When I needed many arena-backed vectors, I wrote an ArenaVec with an API that takes a pointer to the arena in all methods that can allocate, keeping the memory size down.

Yes they do, but since c++11, allocators were not required to be stateless. As a result, containers (or any allocator-aware objects) have needed to hold an instance of the allocator.

I once spent a while building a tool like that. The day I realized that I needed to convert it into something with a top-level loop was a sad one.

.Net now has a non-hierarchical version of this:

http://mattwarren.org/2016/08/16/Preventing-dotNET-Garbage-C...

I believe certain web servers do this, since some memory only has to be kept for the duration of a request.

Nginx does this and it borrowed the idea from Apache.

That's what I do in ACM-ICPC when I use C++! :-)

Reply on news.ycombinator.com