I have a similar image from a ways back, your image gave me pause to go look at what I did to get a such a rendition. Comparing the two, I came to a somewhat counter-intuitive observation…
In rawproc, here’s your image in its “linear” state, before any discretionary processing:
Really, what you’d expect if you were exposing to preserve highlights. There’s probably room to expose more, as the data that fills the histogram is really just that bit of glint in the vase. More on that in a minute…
Here’s my image at the same linear state:
No glinty regions to skew the histogram, but this exposure is definitely “higher”, putting more of the image in the mid-range. I think this is a key consideration to pulling off what needs to happen next…
Before one starts masking regions and yanking them around, IMHO it’s good to see if the tone curve will do the trick. Here’s what I did for mine:
All I did was to pull the bottom part of the curve to the bottom, to crush the background. The increase in slope in the upper part put contrast into the flower, getting it out of the high-key region for no extra effort. I think it’s the “over-exposure” that makes this all work, putting the majority of the data in the workable regions of the histogram.
To treat your image similarly, I had to first add a loggamma curve to lift the shadows out of the well. Then I applied a control-point curve to shape the tones:
The subject didn’t have appreciable light separation from the background, so I had to steepen the curve to pull it up. What you see is as high as I could go before the contrast became too “nasty”, for lack of a better description.
I think my flower had about a stop or so more light than the background, owing to the west-facing patio door that illuminated it more than the rest of the plant back behind the wall. But, I think having the whole image more to the high end of the tone range helped in manipulating the data to enhance that separation.
I know you asked for darktable feedback, but this particular subject and treatment struck a nerve, in a good way…