Add @safe and pure to enforce family

[9rnsr]

By fixing issue 5750, we can add more pure and @safe to enforce family.

[dnadlinger]

It would be great if there wouldn't be so many of these kind of changes, especially since we don't have a style policy on how to format short bodies.

[9rnsr]

OK. I have reverted that indentaitons. It is not my wish to be rejected this change because of the trivial.
But, I dislike not indented code even if they are short bodies.

[dnadlinger]

I wouldn't be in a position to reject your changes anyway, and it's not like I would have a problem with splitting it to two lines, but imho arbitrary style changes for things not covered by the style guide should always be kept at a minimum, since it clutters up diffs and makes code needlessly hard to review.

[jmdavis]

I'd say that rewrites are fine when code is actually being revised, but minor changes to code that isn't being changed for other reasons shouldn't be happening.

[jmdavis]

Well, it does appear to work with arguments to enforce which aren't @safe or pure, which is essentially the same as if the parameter weren't lazy. As long as that behavior is going to stay as-is, then I think that this is a positive change. Even if @safe and pure shouldn't be necessary, they don't actually hurt, and making enforce pure will be a major boon for purity. It's one of the major reasons that stuff in Phobos can't be pure (the other really big one being that Appender can't be used in pure functions).

So, as long as we're sure that the behavior of lazy here is correct and that compiler changes aren't going to cause problems for arguments which aren't really pure or @safe, I think that these changes should be merged in even if @safe and pure shouldn't be necessary.

[9rnsr]

Now issue 6690 was fixed, then we don't need the @safe pure annotations except for the addition to the bailOut.

Which is better, adding @safe pure for descriptive documentation, or removing them with relying on @safe pure inference?

[jmdavis]

I would argue that we should use @safe, pure, and nothrow on templated functions when the intention is that they always have that attribute regardless of what arguments they're instantiated with. It's then clear that they are going to always have that attribute, and it could catch bugs in cases where it somehow becomes possible for them not to have that attribute when inferred. I think that we should leave inferrence for cases where we don't know whether the function will be inferred to have that particular attribute or not.

[jmdavis]

Please fix all of the versions of enforce and its related functions so that line is size_t. I realize that it's not your fault in all cases (like this line), since it was int before in some cases, but it's a bit of a hodgepodge at the moment - including in your changes. It should be size_t in all cases, since __LINE__ is size_t. So, please make the change. Once that's done, I think that we should merge these changes in.

[9rnsr]

OK. I changed them into size_t.

[jmdavis]

Merged.