The problem
The streaming render path has no test coverage, and the reason it has none is a
false negative that reads exactly like a broken feature.
StemDeck draws the waveform lanes two different ways. When the Web Audio engine
owns playback the lanes are an SVG overlay built from peaks. On the streaming
path they are WaveSurfer canvases. The two share nothing: different code,
different bar geometry, different response to a zoom.
Only the first is tested.
Why nobody noticed
Looking for canvases the obvious way finds none:
document.querySelector("#multitrack-container").querySelectorAll("canvas").length
// 0Which says the streaming path renders nothing. It says that after eight seconds
of waiting, with the loading overlay cleared, canplay fired and no failed
requests. It is a convincing result and it is wrong: WaveSurfer renders into
shadow roots, and querySelectorAll does not cross a shadow boundary. Walking
the roots finds nine canvases, correctly sized.
So the cost is not only the missing coverage. It is that the first person to
look concludes the path is broken and either files that, or works around a bug
that does not exist. This already happened once, and the wrong claim reached a
merged pull request body.
What coverage needs to prove
The zoom promise on this path rests on a different mechanism from the SVG one.
WaveSurfer bars are configured in pixels (barWidth 3, barGap 2), so they stay the
same width across a zoom only if it re-renders rather than letting a
fixed-size canvas stretch. That is worth asserting rather than assuming, and it
cannot be asserted without piercing the shadow DOM.
The problem
The streaming render path has no test coverage, and the reason it has none is a
false negative that reads exactly like a broken feature.
StemDeck draws the waveform lanes two different ways. When the Web Audio engine
owns playback the lanes are an SVG overlay built from peaks. On the streaming
path they are WaveSurfer canvases. The two share nothing: different code,
different bar geometry, different response to a zoom.
Only the first is tested.
Why nobody noticed
Looking for canvases the obvious way finds none:
Which says the streaming path renders nothing. It says that after eight seconds
of waiting, with the loading overlay cleared,
canplayfired and no failedrequests. It is a convincing result and it is wrong: WaveSurfer renders into
shadow roots, and
querySelectorAlldoes not cross a shadow boundary. Walkingthe roots finds nine canvases, correctly sized.
So the cost is not only the missing coverage. It is that the first person to
look concludes the path is broken and either files that, or works around a bug
that does not exist. This already happened once, and the wrong claim reached a
merged pull request body.
What coverage needs to prove
The zoom promise on this path rests on a different mechanism from the SVG one.
WaveSurfer bars are configured in pixels (barWidth 3, barGap 2), so they stay the
same width across a zoom only if it re-renders rather than letting a
fixed-size canvas stretch. That is worth asserting rather than assuming, and it
cannot be asserted without piercing the shadow DOM.