RFR: 8000617: It should be possible to allocate memory without the VM dying.

[David Holmes]

On 15/10/2012 9:31 PM, Keith McGuigan wrote:
>
> On Oct 14, 2012, at 7:10 PM, David Holmes wrote:
>
>> On 13/10/2012 1:40 AM, Nils Loodin wrote:
>>> So in general we have two competing wishes here:
>>>
>>> 1. Use std::nothrow, in operator new() and enum in the allocation methods. (basically webrev 03)
>>> Pros: more c++ standard-ish
>>> cons: more boiler plate, not descriptive of what actually happens, two different mechanisms.
>>>
>>>
>>> 2. Use enums everywhere (basically a bugfree version of 04):
>>> pros: more accurate description, cleaner implementation
>>> cons: it's an instance of "not invented here"-syndrom, ie, we're making up our own stuff.
>>
>> I'm less supportive of using an Enum over a boolean now. A two-valued enum is a boolean.
>
> I disagree.  A two-valued enum adds code readability to the code, though I'm sure the resulting machine code is the same.  I'd much rather see a self-describing name at the callsite than a naked 'true' or 'false which requires one to go refer to the function declaration (or definition) to figure out what that value means.   We don't do a good job about this in general in codebase -- it'd be nice to start doing it right.

A boolean constant can achieve the same call-site readability. If the 
parameter is "bool abort_on_oom" then you can have

const bool Alloc::ABORT_ON_OOM = true;
const bool Alloc::RETURN_NULL = false;

But this is a minor and mostly stylistic issue.

David

> --
> - Keith