C++ Logo


Advanced search

Re: [SG10] Teleconference: June 19

From: Gabriel Dos Reis <gdr_at_[hidden]>
Date: Fri, 21 Jun 2013 17:45:16 -0500
"Nelson, Clark" <clark.nelson_at_[hidden]> writes:

| > | The more I look at constexpr, and the draft recommendations, the
| > more I
| > | think that we should be aiming for something simple for chicago.
| > |
| > | Therefore, I'm in favor of just
| > |
| > | __cplusplus_constexpr
| > |
| > | and setting different values to determine what level of language
| > | support is being specified.
| > |
| > | IMHO, if this group starts coming up with 4 macros to detail
| > what
| > | version of constexpr is in play, it's going to be too confusing
| > to use
| > | in practice. What do these macros mean again? Etc.
| >
| > I strongly with what Benjamin said. This suggestion appears to me
| > to
| > be a far superior solutions to alternatives I've seen so far.
| (I assume "agree with" is missing from your first sentence.)

Correct. (I figured it is better for me to write possibly in hurry,
from an uncomfortable place than trying to wait till the comfort
of my office or home. Apologies for unwittingly leaving words out, and
possibly causing confusion.)

| My impression is that Benjamin's comment about "4 macros" was based on a
| misunderstanding about what the document was trying to say. The four names
| in that draft of the document associated with constexpr were just
| alternatives that had been proposed, and from which I was expecting SG10
| to select. It never occurred to me that someone might think I was proposing
| to define four different macros for one feature.
| Benjamin, please correct me if I'm wrong.
| > Furthermore, this should not be done only for constexpr. The
| > approach
| > should be used as a template for all the other macros.
| Benjamin did suggest that several other macros might not be needed,
| including init_captures, generic_lambdas, and decltype_auto. SG10 considered
| them all, and in each case reached the consensus that it would be better to
| keep the specific macro than to try to do it a different way. The general
| consensus was that "fewer macros is better" is not universally true.

conversely, "more macros is better" isn't true either. Please do
consider uses, not just definitions, and their effect on actual software

| I don't know, maybe Benjamin got tired and gave up at that point. :-/ But
| Gaby, if you have any other specific suggestions of macros that you think
| are not necessary, do please speak up.

I think I tried to do that in my previous message; I am still hoping I
won't give up in frustration, like Benjamin. Taking lambdas for
example, I am of the opinion that instead of trying to define macros for
eac single aspect, it is better just to define with values indicating
the date when specific semantics was voted in, or any other values you
would want. Same for decltype.

-- Gaby

Received on 2013-06-22 00:45:32