adding zeroize for stream_core.rs (#1494) #1495
Conversation
…ypto#1501) Replaces the old, significantly less flexible `LinearCombination` trait with the newer `LinearCombinationExt` trait, removing the old trait completely and renaming the new one. The bounds for `CurveArithmetic::ProjectivePoint` have also been updated accordingly.
| @@ -35,7 +38,7 @@ pub trait StreamClosure: BlockSizeUser { | |||
| } | |||
|
|
|||
| /// Block-level synchronous stream ciphers. | |||
| pub trait StreamCipherCore: BlockSizeUser + Sized { | |||
| pub trait StreamCipherCore: BlockSizeUser + ParBlocksSizeUser + Sized { | |||
There was a problem hiding this comment.
As I wrote in the issue, this approach is incorrect. StreamCipherCore does not "know" at compile time the number of blocks which can be generated in parallel at runtime. It's known only by backends which can be selected using runtime target feature detection. For example, chacha20 has SSE2 backend which generates only one block and AVX2 which generates 4 blocks. By default on x86 targets the crate uses runtime autodetection to select between those.
There was a problem hiding this comment.
The approach is not ideal, but it would still work if the x86 ParBlocksSize for ChaChaCore was set to U4.
Is there a better approach for choosing the size at runtime? The last time I tried to pick a variable-sized [T; N] at runtime, the compiler was yelling at me because it did not know N at compile time. Is this possible with the array crate, or something similar?
Regarding #1494