Mention that examples and doctests require std

[vks]

Also remove pointless attributes. (#![cfg(feature="std")] does not work
for binaries, because main will be missing.)

[dhardy]

Are you sure it's pointless? I'm not really sure why cargo test --tests --no-default-features fails sometimes.

[vks]

Are you sure it's pointless?

You are right, the CI fails due to this patch. Not sure why. Maybe the cargo behavior is inconsistent across Rust versions?

[pitdicker]

Can you think of another place to add the comment? I just removed that section from the readme in #437...

[pitdicker]

The difference is that some travis configurations run

cargo test --features serde1,log,nightly,alloc

instead of the more complete:

cargo test --tests --features serde1,log,nightly,alloc

We could also use --lib and --examples if we want to be explicit.

[vks]

@dhardy
The CI fails on only two platforms and that was the only explanation I could think of. But according to the CI the cargo versions are the same. I don't understand those failures.

@pitdicker

Can you think of another place to add the comment? I just removed that section from the readme in #437...

I'm opposed to removing the testing section if we decide against making the examples and doctests compile without std. I probably will not be the last one to try to run cargo test --no-default-features. This is important to know for any contributor to Rand.

The difference is that some travis configurations run cargo test --features serde1,log,nightly,alloc instead of the more complete cargo test --tests --features serde1,log,nightly,alloc

Are you sure this is the reason? AFAIK, cargo test --features serde1,log,nightly,alloc does not disable std, so I don't see how this would be related to the CI failures.

[pitdicker]

Are you sure this is the reason?

Yes, because one also tests the examples, the other not.

[vks]

The documentation bits of this are superseded by #443. I think the only thing left is cleaning up the examples / CI scripts.

[dhardy]

Essentially this is a compromise between code simplicity/clarity and making the tests work. These changes to the examples require a few tweaks to scripts e.g.:

cross test --tests --no-default-features --target mips-unknown-linux-gnu

to (I guess):

cross test --lib --no-default-features --target mips-unknown-linux-gnu

But I'm not sure it's worth picking a different compromise than what we already have?

[dhardy]

So it sounds like there's no point making these changes.