refactor(sdk): Allow to send waveform for any audio message

[zecakeh]

We plan to use this in Fractal to always be able to present a waveform for an audio message.

The second commit might be more controversial: it changes the format of the waveform from a list of u16 between 0 and 1024 to a list of f32 between 0 and 1. This is done because the value between 0 and 1024 used in the event is quite arbitrary (and they have changed in MSC3246 since then), most clients should end up with values between 0 and 1, and need to convert it for sending it. So this centralizes this conversion in the SDK.

[codspeed-hq]

CodSpeed Performance Report

Merging #5732 will not alter performance

_{Comparing zecakeh:audio-waveform (36f147a) with main (588d604)}

Summary

✅ 50 untouched

[codecov]

Codecov Report

❌ Patch coverage is 0% with 5 lines in your changes missing coverage. Please review.
✅ Project coverage is 88.42%. Comparing base (588d604) to head (36f147a).
⚠️ Report is 3 commits behind head on main.

Files with missing lines	Patch %	Lines
crates/matrix-sdk/src/attachment.rs	0.00%	4 Missing ⚠️
crates/matrix-sdk/src/room/mod.rs	0.00%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #5732      +/-   ##
==========================================
- Coverage   88.43%   88.42%   -0.01%     
==========================================
  Files         360      360              
  Lines       99796    99794       -2     
  Branches    99796    99794       -2     
==========================================
- Hits        88250    88245       -5     
- Misses       7405     7409       +4     
+ Partials     4141     4140       -1     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[Hywan]

The changes look good to me. I'm a bit annoyed by the regression with the waveform format, but since it's a different type and it's documented in the CHANGELOG, I think it's fine.

[Hywan]

Re-running the benchmarks. I suspect the result is misleading and could be due to noise.

@zecakeh can you fix the conflicts please?

[bnjbvr]

Thanks, we're on board with this change, and as you said it'll make the switch to the new MSC simpler. We might delay the landing of this PR, since we want to limit landing breaking changes at the FFI layer until the Matrix conference.

One comment I would have is: why to use f32 and not f64? We'd have a much better precision when it comes to the [0; 1] range, leading to a better precision in the normalized range; and I don't think we're getting any notable speedup for using 32 bits instead of 64 bits here.

Would it make sense to change to a f64 in this patch?

[zecakeh]

I don't mind changing to f64 but I chose f32 because it is precise enough to be converted to an integer value between 0 and 1024 and it should be enough to draw it too so I don't see what we gain by increasing the precision?

[bnjbvr]

I wary of using float32s, because they can't represent finely small values, viz. values in the [0; 1] range. That is, many small float32 values would be normalized to the same int in the [0; 1024] range, so if you looked at a flat projection of the f32 range to the int10 range, you might see holes in the int range. Since float64 have so much more precision, it's unlikely that this problem would happen with those.

[zecakeh]

I am not convinced that this is true, because f32 has a precision of 6 significant digits, and for integer values between 0 and 1024 we only need 4.

[bnjbvr]

I've double-checked that with a program, and agree with your conclusion; good point.

Approving, but we've been requested to not merge this immediately.

[zecakeh]

No problem, I'll rebase this after the Matrix conference then to solve the conflicts.

[poljar]

No problem, I'll rebase this after the Matrix conference then to solve the conflicts.

The EX releases for the conf have been done, you can now rebase so we can merge.

[zecakeh]

The EX releases for the conf have been done, you can now rebase so we can merge.

Done

[zecakeh]

Oh I forgot to fix the typo.

[zecakeh]

Done.