Make getrandom a non-pub dep; rename feature to os_rng

[dhardy]

• Added a CHANGELOG.md entry

Remove getrandom from the public API so that upgrading the version used will be non-breaking.

Rename the feature getrandom to os_rng. This is in line with small_rng, std_rng, and its effect of enabling OsRng, from_os_rng and rand::rng.

[newpavlov]

I think getrandom::Error can be still exposed in the public API of rand_core. I don't think it's worth to wrap it in a custom type at this stage.

[dhardy]

I think getrandom::Error can be still exposed in the public API of rand_core. I don't think it's worth to wrap it in a custom type at this stage.

I disagree: getrandom is an implementation detail that does not need to be exposed to the user. Bumping the getrandom version should not require a version bump to rand_core.

It is possible to add a method like fn as_getrandom_0_2(self) -> Option<getrandom::Error>, but it would need to be feature-gated behind a getrandom_0_2 feature I think. Or use an external crate which provides traits like raw-window-handle (but you still end up needing the version-specific feature in rand_core).

For now I'm not sure it's worth bothering with this however.

[newpavlov]

I am fine with wrapping the error, it's just that I don't see much merit in it since it's unlikely that API of getrandom will evolve faster than API of rand. It would've been a different story if it was a rand v1.0 release.

[dhardy]

It would've been a different story if it was a rand v1.0 release.

Pretty much all the changes I've been planning are already in master, so my plan is to essentially treat v0.9 as an extended Beta for v1.0.

Yes, there's #1261, but if any of that lands it's more likely to be in v2.0 or later. Much of it depends on future Rust features.

[dhardy]

• Added a CHANGELOG.md entry

Remove getrandom from the public API so that upgrading the version used will be non-breaking.

Rename the feature getrandom to os_rng. This is in line with small_rng, std_rng, and its effect of enabling OsRng, from_os_rng and rand::rng.

[newpavlov]

I think getrandom::Error can be still exposed in the public API of rand_core. I don't think it's worth to wrap it in a custom type at this stage.

[dhardy]

I think getrandom::Error can be still exposed in the public API of rand_core. I don't think it's worth to wrap it in a custom type at this stage.

I disagree: getrandom is an implementation detail that does not need to be exposed to the user. Bumping the getrandom version should not require a version bump to rand_core.

It is possible to add a method like fn as_getrandom_0_2(self) -> Option<getrandom::Error>, but it would need to be feature-gated behind a getrandom_0_2 feature I think. Or use an external crate which provides traits like raw-window-handle (but you still end up needing the version-specific feature in rand_core).

For now I'm not sure it's worth bothering with this however.

[newpavlov]

I am fine with wrapping the error, it's just that I don't see much merit in it since it's unlikely that API of getrandom will evolve faster than API of rand. It would've been a different story if it was a rand v1.0 release.

[dhardy]

It would've been a different story if it was a rand v1.0 release.

Pretty much all the changes I've been planning are already in master, so my plan is to essentially treat v0.9 as an extended Beta for v1.0.

Yes, there's #1261, but if any of that lands it's more likely to be in v2.0 or later. Much of it depends on future Rust features.

[dhardy]

• Added a CHANGELOG.md entry

Remove getrandom from the public API so that upgrading the version used will be non-breaking.

Rename the feature getrandom to os_rng. This is in line with small_rng, std_rng, and its effect of enabling OsRng, from_os_rng and rand::rng.

[newpavlov]

I think getrandom::Error can be still exposed in the public API of rand_core. I don't think it's worth to wrap it in a custom type at this stage.

[dhardy]

I think getrandom::Error can be still exposed in the public API of rand_core. I don't think it's worth to wrap it in a custom type at this stage.

I disagree: getrandom is an implementation detail that does not need to be exposed to the user. Bumping the getrandom version should not require a version bump to rand_core.

It is possible to add a method like fn as_getrandom_0_2(self) -> Option<getrandom::Error>, but it would need to be feature-gated behind a getrandom_0_2 feature I think. Or use an external crate which provides traits like raw-window-handle (but you still end up needing the version-specific feature in rand_core).

For now I'm not sure it's worth bothering with this however.

[newpavlov]

I am fine with wrapping the error, it's just that I don't see much merit in it since it's unlikely that API of getrandom will evolve faster than API of rand. It would've been a different story if it was a rand v1.0 release.

[dhardy]

It would've been a different story if it was a rand v1.0 release.

Pretty much all the changes I've been planning are already in master, so my plan is to essentially treat v0.9 as an extended Beta for v1.0.

Yes, there's #1261, but if any of that lands it's more likely to be in v2.0 or later. Much of it depends on future Rust features.

[dhardy]

• Added a CHANGELOG.md entry

Remove getrandom from the public API so that upgrading the version used will be non-breaking.

Rename the feature getrandom to os_rng. This is in line with small_rng, std_rng, and its effect of enabling OsRng, from_os_rng and rand::rng.

[newpavlov]

I think getrandom::Error can be still exposed in the public API of rand_core. I don't think it's worth to wrap it in a custom type at this stage.

[dhardy]

I think getrandom::Error can be still exposed in the public API of rand_core. I don't think it's worth to wrap it in a custom type at this stage.

I disagree: getrandom is an implementation detail that does not need to be exposed to the user. Bumping the getrandom version should not require a version bump to rand_core.

It is possible to add a method like fn as_getrandom_0_2(self) -> Option<getrandom::Error>, but it would need to be feature-gated behind a getrandom_0_2 feature I think. Or use an external crate which provides traits like raw-window-handle (but you still end up needing the version-specific feature in rand_core).

For now I'm not sure it's worth bothering with this however.

[newpavlov]

I am fine with wrapping the error, it's just that I don't see much merit in it since it's unlikely that API of getrandom will evolve faster than API of rand. It would've been a different story if it was a rand v1.0 release.

[dhardy]

It would've been a different story if it was a rand v1.0 release.

Pretty much all the changes I've been planning are already in master, so my plan is to essentially treat v0.9 as an extended Beta for v1.0.

Yes, there's #1261, but if any of that lands it's more likely to be in v2.0 or later. Much of it depends on future Rust features.

[dhardy]

• Added a CHANGELOG.md entry

Remove getrandom from the public API so that upgrading the version used will be non-breaking.

Rename the feature getrandom to os_rng. This is in line with small_rng, std_rng, and its effect of enabling OsRng, from_os_rng and rand::rng.

[newpavlov]

I think getrandom::Error can be still exposed in the public API of rand_core. I don't think it's worth to wrap it in a custom type at this stage.

[dhardy]

I think getrandom::Error can be still exposed in the public API of rand_core. I don't think it's worth to wrap it in a custom type at this stage.

I disagree: getrandom is an implementation detail that does not need to be exposed to the user. Bumping the getrandom version should not require a version bump to rand_core.

It is possible to add a method like fn as_getrandom_0_2(self) -> Option<getrandom::Error>, but it would need to be feature-gated behind a getrandom_0_2 feature I think. Or use an external crate which provides traits like raw-window-handle (but you still end up needing the version-specific feature in rand_core).

For now I'm not sure it's worth bothering with this however.

[newpavlov]

I am fine with wrapping the error, it's just that I don't see much merit in it since it's unlikely that API of getrandom will evolve faster than API of rand. It would've been a different story if it was a rand v1.0 release.

[dhardy]

It would've been a different story if it was a rand v1.0 release.

Pretty much all the changes I've been planning are already in master, so my plan is to essentially treat v0.9 as an extended Beta for v1.0.

Yes, there's #1261, but if any of that lands it's more likely to be in v2.0 or later. Much of it depends on future Rust features.

[dhardy]

• Added a CHANGELOG.md entry

Remove getrandom from the public API so that upgrading the version used will be non-breaking.

Rename the feature getrandom to os_rng. This is in line with small_rng, std_rng, and its effect of enabling OsRng, from_os_rng and rand::rng.

[newpavlov]

I think getrandom::Error can be still exposed in the public API of rand_core. I don't think it's worth to wrap it in a custom type at this stage.

[dhardy]

I think getrandom::Error can be still exposed in the public API of rand_core. I don't think it's worth to wrap it in a custom type at this stage.

I disagree: getrandom is an implementation detail that does not need to be exposed to the user. Bumping the getrandom version should not require a version bump to rand_core.

It is possible to add a method like fn as_getrandom_0_2(self) -> Option<getrandom::Error>, but it would need to be feature-gated behind a getrandom_0_2 feature I think. Or use an external crate which provides traits like raw-window-handle (but you still end up needing the version-specific feature in rand_core).

For now I'm not sure it's worth bothering with this however.

[newpavlov]

I am fine with wrapping the error, it's just that I don't see much merit in it since it's unlikely that API of getrandom will evolve faster than API of rand. It would've been a different story if it was a rand v1.0 release.

[dhardy]

It would've been a different story if it was a rand v1.0 release.

Pretty much all the changes I've been planning are already in master, so my plan is to essentially treat v0.9 as an extended Beta for v1.0.

Yes, there's #1261, but if any of that lands it's more likely to be in v2.0 or later. Much of it depends on future Rust features.

[dhardy]

• Added a CHANGELOG.md entry

Remove getrandom from the public API so that upgrading the version used will be non-breaking.

Rename the feature getrandom to os_rng. This is in line with small_rng, std_rng, and its effect of enabling OsRng, from_os_rng and rand::rng.

[newpavlov]

I think getrandom::Error can be still exposed in the public API of rand_core. I don't think it's worth to wrap it in a custom type at this stage.

[dhardy]

I think getrandom::Error can be still exposed in the public API of rand_core. I don't think it's worth to wrap it in a custom type at this stage.

I disagree: getrandom is an implementation detail that does not need to be exposed to the user. Bumping the getrandom version should not require a version bump to rand_core.

It is possible to add a method like fn as_getrandom_0_2(self) -> Option<getrandom::Error>, but it would need to be feature-gated behind a getrandom_0_2 feature I think. Or use an external crate which provides traits like raw-window-handle (but you still end up needing the version-specific feature in rand_core).

For now I'm not sure it's worth bothering with this however.

[newpavlov]

I am fine with wrapping the error, it's just that I don't see much merit in it since it's unlikely that API of getrandom will evolve faster than API of rand. It would've been a different story if it was a rand v1.0 release.

[dhardy]

It would've been a different story if it was a rand v1.0 release.

Pretty much all the changes I've been planning are already in master, so my plan is to essentially treat v0.9 as an extended Beta for v1.0.

Yes, there's #1261, but if any of that lands it's more likely to be in v2.0 or later. Much of it depends on future Rust features.

[dhardy]

• Added a CHANGELOG.md entry

Remove getrandom from the public API so that upgrading the version used will be non-breaking.

Rename the feature getrandom to os_rng. This is in line with small_rng, std_rng, and its effect of enabling OsRng, from_os_rng and rand::rng.

[newpavlov]

I think getrandom::Error can be still exposed in the public API of rand_core. I don't think it's worth to wrap it in a custom type at this stage.

[dhardy]

I think getrandom::Error can be still exposed in the public API of rand_core. I don't think it's worth to wrap it in a custom type at this stage.

I disagree: getrandom is an implementation detail that does not need to be exposed to the user. Bumping the getrandom version should not require a version bump to rand_core.

It is possible to add a method like fn as_getrandom_0_2(self) -> Option<getrandom::Error>, but it would need to be feature-gated behind a getrandom_0_2 feature I think. Or use an external crate which provides traits like raw-window-handle (but you still end up needing the version-specific feature in rand_core).

For now I'm not sure it's worth bothering with this however.

[newpavlov]

I am fine with wrapping the error, it's just that I don't see much merit in it since it's unlikely that API of getrandom will evolve faster than API of rand. It would've been a different story if it was a rand v1.0 release.

[dhardy]

It would've been a different story if it was a rand v1.0 release.

Pretty much all the changes I've been planning are already in master, so my plan is to essentially treat v0.9 as an extended Beta for v1.0.

Yes, there's #1261, but if any of that lands it's more likely to be in v2.0 or later. Much of it depends on future Rust features.