perf: Remove anyhow from P2P code

[eichhorl]

anyhow captures a backtrace whenever an error is created. This can be quite expensive and has lead to a performance regression after updating the libunwind version in #4848.

The P2P code does not make use of this backtrace anywhere.

Therefore, this PR removes anyhow from all P2P crates, and replaces it with a custom opaque error type, that doesn't capture backtraces.

See also this issue: dtolnay/anyhow#346

[kpop-dfinity]

anyhow captures a backtrace whenever an error is created. This can be quite expensive and lead to a performance regression after updating the libunwind version in #4848.

Does it make a noticeable difference in our performance tests?

[eichhorl]

anyhow captures a backtrace whenever an error is created. This can be quite expensive and lead to a performance regression after updating the libunwind version in #4848.

Does it make a noticeable difference in our performance tests?

I only tried the tecdsa_performance_test, but there I observed an increase of signatures per second from ~1.77/s to ~1.97/s, and an increase in finalization rate from ~1/s to ~2/s.

[kpop-dfinity]

anyhow captures a backtrace whenever an error is created. This can be quite expensive and lead to a performance regression after updating the libunwind version in #4848.

Does it make a noticeable difference in our performance tests?

I only tried the tecdsa_performance_test, but there I observed an increase of signatures per second from ~1.77/s to ~1.97/s, and an increase in finalization rate from ~1/s to ~2/s.

wow

[kpop-dfinity]

Nice catch!