Classes on stack

[Steven Schveighoffer]

On 9/2/22 3:54 PM, Walter Bright wrote:
> On 9/2/2022 6:23 AM, Steven Schveighoffer wrote:
>> What do you mean? If you are suggesting that probably the module 
>> sorting is insignificant, it definitely is not very intensive.
> 
> I am indeed suggesting it is insignificant.

I'd hazard to guess it's indistinguishable really. Probably on the order 
of a couple milliseconds, even on a large module set.

It's not so much the performance, but the principle, and also the 
requirement of code to do something that is knowable really at link-time.

> 
> In the olden days, I (and others) would just have the executable 
> self-patch itself. But such behavior was then banned by the OS.
> 

I'm thinking more about either an intermediate step to the linker (once 
all objects are known), or a post-link step. Not something that happens 
at runtime.

What *could* happen is some --DRT switch to output the correct sorting, 
and then another tool that takes that as input, and can then properly 
edit the executable, or provide a specialized file to the compiler so it 
spits out a pre-organized module list. Can they be weak symbols, and 
then you just link the object file with that data ahead of everything 
else? Or, just the rt_init function could be told "use this pointer 
array as the list instead of the sorting."

Actually, I think that last idea is doable without any compiler 
changes... I'll have to investigate.

-Steve