Restrict length

[andralex]

Fixes https://issues.dlang.org/show_bug.cgi?id=16566. hasLength was in fact intended to allow only integrals, but its code predates alias this and we didn't update it.

Throughout Phobos we use interchangeably auto s = r.length;, size_t s = r.length;, cache the length of a range in a variable, or just use it as a size_t. Supporting other types than size_t meaningfully would put too much burden on the definition and the implementation.

[jmdavis]

Shouldn't the restriction be that it be size_t only and not ulong? We even removed the ability for iota to have a length of ulong (unless size_t is ulong) because of the problems that it caused.

[burner]

the doc unittest could use some more failing asserts

[JackStouffer]

Ping @andralex

[9il]

ping @andralex

[andralex]

I plan to be more aggressive about this: if an artifact defines length to have type ulong on a 32-bit system, I'll yield a deprecation message. @MartinNowak does this work for you?

[JackStouffer]

Ping @MartinNowak

[wilzbach]

The XREF macro was deprecated, please use the universal REF, e.g. `$(REF, byCodeUnit, std, utf)

[wilzbach]

I plan to be more aggressive about this: if an artifact defines length to have type ulong on a 32-bit system, I'll yield a deprecation message.

How about hacking template constraints together for a pseudo-deprecation message as we deprecated on templates doesn't seem to work:

private enum bool oldHasLengthImpl(R) = !isNarrowString!R && is(typeof(
      (inout int = 0)
      {
          R r = R.init;
          ulong l = r.length;
      }));

private enum bool newHasLengthImpl(R) = !isNarrowString!R && is(typeof(
      (inout int = 0)
      {
          R r = R.init;
          auto x = r.length;
          static assert(is(typeof(x) == size_t) || is(typeof(x) == ulong));
      }));

template hasLength(R)
if (oldHasLengthImpl!R && newHasLengthImpl!R)
{
    enum bool hasLength = true;
}

template hasLength(R)
if (oldHasLengthImpl!R && !newHasLengthImpl!R)
{
    pragma(msg, "Deprecation: length of " ~ R.stringof ~ " needs to return size_t or ulong");
    enum bool hasLength = true;
}

template hasLength(R)
if (!oldHasLengthImpl!R)
{
    enum bool hasLength = false;
}

In Phobos there is only this one violation in std.uni.simpleCaseFoldings

[quickfur]

ping @andralex

[andralex]

OK, this is reviewable now. The new solution is nicer than the inout trick and uses a deprecation step.

[wilzbach]

You should probably include an @@@deprecated... comment and deprecation end date here, so that one can easily grep for it and remove this bit eventually.

[dnadlinger]

I'd rather not imitate the compiler's message format, as it might be confusing for people compiling with deprecation warnings as errors, or who have them disabled entirely. LGTM otherwise.

[andralex]

Updated

[wilzbach]

Any reason why you mix the Length == size_t and Length : ulong styles?

[andralex]

These aren't styles. == checks for exact match, : is convertibility.

[wilzbach]

Shouldn't this be updated as well if you deprecate ulong?

[andralex]

oh ok will update text

[andralex]

done

[dlang-bot]

Fix	Bugzilla	Description
✓	16566	hasLength should enforce that length has type size_t

[wilzbach]

@andralex I changed your commit message, s.t. the bug can detect the issue and added a small changelog entry (basically just copied your last paragraph).

[wilzbach]

Let's get this shipped ;-)

[MartinNowak]

How is the FILE and LINE of std.range.primitives interesting here? You'd need to pass them as defaults for template arguments to get the instantiating side.

[andralex]

They emulate what the deprecated facility does.