feat!: unite alphabet, join, padding

[ChALkeR]

Depends on:

• feat!: use Uint8Array in alphabet() and radix #50

Main:

==== base32 ====
base32 (encode) 32 B scure x 715,819 ops/sec @ 1μs/op ± 1.30% (1μs..5ms)
base32 (decode) 32 B scure x 1,273,885 ops/sec @ 785ns/op ± 2.69% (583ns..7ms)
base32 (encode) 64 B scure x 408,663 ops/sec @ 2μs/op
base32 (decode) 64 B scure x 753,012 ops/sec @ 1μs/op ± 1.09% (1μs..1ms)
base32 (encode) 1 KB scure x 26,949 ops/sec @ 37μs/op ± 1.72% (35μs..8ms)
base32 (decode) 1 KB scure x 50,607 ops/sec @ 19μs/op
base32 (encode) 8 KB scure x 3,285 ops/sec @ 304μs/op
base32 (decode) 8 KB scure x 5,789 ops/sec @ 172μs/op
base32 (encode) 1 MB scure x 13 ops/sec @ 74ms/op ± 2.81% (69ms..80ms)
base32 (decode) 1 MB scure x 25 ops/sec @ 39ms/op ± 3.51% (36ms..50ms)

This branch:

==== base32 ====
base32 (encode) 32 B scure x 1,001,001 ops/sec @ 999ns/op ± 1.57% (791ns..3ms)
base32 (decode) 32 B scure x 1,908,396 ops/sec @ 524ns/op ± 4.44% (333ns..8ms)
base32 (encode) 64 B scure x 579,374 ops/sec @ 1μs/op ± 1.27% (1μs..1ms)
base32 (decode) 64 B scure x 1,019,367 ops/sec @ 981ns/op ± 4.69% (791ns..15ms)
base32 (encode) 1 KB scure x 39,094 ops/sec @ 25μs/op
base32 (decode) 1 KB scure x 94,277 ops/sec @ 10μs/op
base32 (encode) 8 KB scure x 5,125 ops/sec @ 195μs/op
base32 (decode) 8 KB scure x 12,137 ops/sec @ 82μs/op
base32 (encode) 1 MB scure x 26 ops/sec @ 38ms/op ± 3.63% (35ms..47ms)
base32 (decode) 1 MB scure x 93 ops/sec @ 10ms/op

This can be further (significantly) optimized with TextEncoder/TextDecoder if available and not a polyfill

Also, cloc reports 519 lines of code in index.ts on this branch, but 563 in main
This simplifies things

[paulmillr]

note that alphabet is used here:

https://github.com/paulmillr/scure-bip39/blob/753ea75ac2621b71b64e7899bd72cfd140107296/src/index.ts#L54-L65

we can probably make a specific version for bip39, if the perf benefits are really good, but backwards compat is better be good

[paulmillr]

This can be further (significantly) optimized with TextEncoder/TextDecoder if available and not a polyfill

Can we use it when it's available and fallback when it's not?

[ChALkeR]

@paulmillr yes, that'll be just a few extra lines of code
I'll file that in a separate PR on top of this

[paulmillr]

overall utils export (alphabet etc) is non-core functionality so i'm leaning towards just breaking it. 2.0.0 was only recently released, which means very few people upgraded. Also scure-bip39 uses 2.0.0 of scure-base, which means we can use separate bip39-specific alphabet function in 2.1.0.

I've also searched across github and only 2 projects seem to be using utils:

• https://github.com/polkadot-js/build-deno.land/blob/b8c970aecd1e98d0cd907e0b2bee5bf579b7595e/util-crypto/base32/bs32.ts#L8
• https://github.com/kahnpoint/arc/blob/c7befa4babccd9996dc8691a6759cb07f5f87b98/src/b32.ts#L8

[ChALkeR]

@paulmillr this would still be a semver-major though
Also for radix and bech reasons

[paulmillr]

@ChALkeR radix is also non-core functionality.

Can we leave bech inputs as-is? This would allow it to be released as 2.1.

[paulmillr]

see this https://www.npmjs.com/package/@scure/base?activeTab=versions - i'd argue that for non-core functionality, pragmatically, we can break stuff for 1% of weekly users.

"ideally" v3 would've been released for this change but it's painful to not have synced scure packages (base, bip32, bip39, btc-signer)

[paulmillr]

another reason not to break bech is the fact that it's using 5-bit words and NOT real uint8 stuff.

[paulmillr]

Thanks for the speed-up. Not planned for now, but we'll re-evaluate the architecture for v3.