This enhancement focuses on improving the usability of the http crate when integrating Rust HTTP libraries with other VM engines, such as V8 (specifically for, but not limited to Deno). The goal is to provide embedding users with a specific method that allows extracting a static 7-bit string from a well-known ASCII header name. By treating header strings as interned strings in scripting engines [1][2], we can effectively reduce garbage collection pressure, leading to enhanced performance and efficiency.
To achieve this, we propose introducing an Option<&'static [u8]> method for header names. This approach offers a more efficient means of passing header names across embedding boundaries. When working with static ASCII header names, this method allows direct access to the header name bytes across embedding boundaries, without the need for additional memory allocation or garbage collection overhead.
This may also be useful for Python embedders [2], as (I believe) the CPython engine allows creation of strings from const 1-byte buffers.
Alternatives
While we do have an as_bytes method on a header name, this does not tell us the provenance of those bytes, and whether the header is a StandardHeader. We cannot use these as "externally interned" strings in an embedding context because they may be allocated.
The hash function of the header could be used allow us to map between standard headers and interned strings in the embedded engine, but this requires an additional string comparison operation.
Description
This enhancement focuses on improving the usability of the
httpcrate when integrating Rust HTTP libraries with other VM engines, such as V8 (specifically for, but not limited to Deno). The goal is to provide embedding users with a specific method that allows extracting a static 7-bit string from a well-known ASCII header name. By treating header strings as interned strings in scripting engines[1][2], we can effectively reduce garbage collection pressure, leading to enhanced performance and efficiency.To achieve this, we propose introducing an
Option<&'static [u8]>method for header names. This approach offers a more efficient means of passing header names across embedding boundaries. When working with static ASCII header names, this method allows direct access to the header name bytes across embedding boundaries, without the need for additional memory allocation or garbage collection overhead.This may also be useful for Python embedders
[2], as (I believe) the CPython engine allows creation of strings from const 1-byte buffers.Alternatives
While we do have an
as_bytesmethod on a header name, this does not tell us the provenance of those bytes, and whether the header is aStandardHeader. We cannot use these as "externally interned" strings in an embedding context because they may be allocated.The hash function of the header could be used allow us to map between standard headers and interned strings in the embedded engine, but this requires an additional string comparison operation.
References
[1]V8'sNewExternalOneBytemethod:https://v8docs.nodesource.com/node-15.0/d2/db3/classv8_1_1_string.html#a2b9afa662981b864fdadbc4fe1eb7480
[2]Python'sPyUnicode_FromKindAndDatamethod (which I believe will avoid copies where necessary):https://docs.python.org/3/c-api/unicode.html#c.PyUnicode_FromKindAndData