On Thursday, 16 April 2026 09:07:58 Pacific Daylight Time Evan Teran via Std-
Proposals wrote:
> @Thiago, I would propose that it throw a length_error as that's what I
> believe the behavior would have been, had size_type been signed originally.
>
> I can't say that I agree with you because if a signed integer API for this
> were introduced, it would have to check for negative values regardless.
> Unless I misunderstood that part of the objection.

Making a throwing API would make people use the unsigned API even with a
signed calculation just to avoid the runtime check for the throw. If anything,
just make it a precondition.

But maybe don't: why is the value negative in the first place? It's often
because a calculation unexpectedly underflowed. If reserve(0) would effectively
just be shrink_to_fit(), why would reserve(-1) be any different?

What's more, what is the prior art? The Qt containers that do use signed types
for indices and sizes have a prior behaviour: reserve(-1) is the same as
reserve(0), which is squeeze() / shrink_to_fit().

-- 
Thiago Macieira - thiago (AT) macieira.info - thiago (AT) kde.org
  Principal Engineer - Intel Data Center - Platform & Sys. Eng.