fix Issue 18963 - Relax restrictions on 'return' parameters when para…

[WalterBright]

…meter is not a pointer

[dlang-bot]

Thanks for your pull request, @WalterBright!

Bugzilla references

Auto-close	Bugzilla	Severity	Description
✓	18963	enhancement	Relax restrictions on 'return' parameters when parameter is not a pointer

Testing this PR locally

If you don't have a local development environment setup, you can use Digger to test this PR:

dub fetch digger
dub run digger -- build "master + dmd#8346"

[WalterBright]

This is blocking dlang/phobos#6560

[WalterBright]

@JinShil the spec doesn't mention return parameters at all.

[JinShil]

the spec doesn't mention return parameters at all.

It mentions return ref parameters. But the lack of documentation is all the more reason to document it now, and through that documentation justify this PR.

[schveiguy]

Just want to voice my support here. I don't have enough experience to review this, to know that it's correctly implemented. But speaking from experience with inout (I originally wanted inout to have similar errors like "can't have inout return without inout parameters" and it's horrible for generic code).

[WalterBright]

But the lack of documentation is all the more reason to document it now, and through that documentation justify this PR.

1. Dip1000 is not in the spec. It certainly needs to be done, but I'm pretty reluctant to put the entire Phobos implementation on hold for another year because of this. Making everything serial instead of parallel is not productive.

2. I can't get even simple PRs to the spec pulled. Last January, I resolved to go through and make the spec more accurate and more formal. But this all ground to a halt due to complete disinterest in anyone reviewing or pulling them.

3. This PR is a trivial and obvious improvement. The proposed behavior is the same as for scope which is already implemented.

4. This PR is not a dot i and cross t thing. It's blocking valuable and important work.

5. There's an Enhancement Request referenced which describes it.

[JinShil]

Dip1000 is not in the spec. It certainly needs to be done, but I'm pretty reluctant to put the entire Phobos implementation on hold for another year because of this. Making everything serial instead of parallel is not productive.

Too often, features are added to D without sufficient documentation, and DIP1000 is the worst offender. I'm not asking that you remedy that entire problem right now, I'd just like to see a section on Return Parameters with a brief explanation of it's purpose, a bullet point calling out this PR's exception, and perhaps an example. I don't think that's too much to ask.

I can't get even simple PRs to the spec pulled. Last January, I resolved to go through and make the spec more accurate and more formal. But this all ground to a halt due to complete disinterest in anyone reviewing or pulling them.

Indeed, and I'm sorry about that. There's a lot of disinterest in the things that I and others care about too. But it's not just disinterest. Many of us are not capable of certifying certain PRs. I recommend you put a time limit on those PRs that aren't getting attention. If noone raises any objections after a week, merge it. Noone has a right to complain if they chose not to participate, and we can always submit corrections if errors are found after the fact.

This PR is not a dot i and cross t thing. It's blocking valuable and important work.

The sooner you add the docs, the sooner you'll be unblocked.

There's an Enhancement Request referenced which describes it.

Bugzilla is not a substitute for proper documentation.

[andralex]

Please don't let bureaucracy smother progress. Thanks.