Fix #682

[dhardy]

No description provided.

[dhardy]

I also included the switch to run-time failure for WASM, as mentioned in #678 and #681.

@newpavlov would you please review?

[vks]

This adds a new default feature, so this is a breaking change. This warrants increasing the Rand version to 0.7.

[dhardy]

Why is that a breaking change? Besides, this was added in 0.6.2; this is just documenting it better.

BTW I had to push yet another fix to the changelog and prepare the next rand_os release (perhaps a new PR would have been better)

[vks]

@dhardy Adding a new default feature that gates existing functionality breaks builds with --no-default-features (see rust-num/num#297).

[dhardy]

I don't believe this is a problem since it is implied by std and the functionality was already feature-gated behind std.

It's actually pointless as a feature flag since it is only usable exactly when std is usable, but Cargo makes all optional dependencies into "features", hence I thought we should document it.

[vks]

I don't believe this is a problem since it is implied by std and the functionality was already feature-gated behind std.

I think you are right, this should be no problem.

[newpavlov]

As a patch for v0.6 WASM issues looks good to me! Though I still strongly believe that rand_os should use compile-time errors instead of the runtime ones. I guess we can discuss it in a separate issue.

[newpavlov]

Does this module have to be public? Can't it be pub(crate)?

[dhardy]

I don't know. The module isn't required at all for building and we don't have a running test-suite. However it's not referenced within the crate, so I don't think pub(crate) is appropriate.

[dhardy]

Though I still strongly believe that rand_os should use compile-time errors instead of the runtime ones.

Ideally, yes, but in this that appears to be difficult, and likely not worth the effort of figuring out the problems other people have as a result.