The main problem is certainly given that we humans have our attention to brightness values more in the middle (to what ever our eyes adjust). The darker and brighter tones are not really as important. The funny thing was, that CRT tubes (super simplified) work in a similar fashion, minus the “aperture” of our eyes: It had a similar “Gamma”.
Over a decade ago, Stu Maschwitz (Co founder of the Studio: The Orphanage, San Francisco, CA) started to complain about Gamma and fought for a “Linear Light Work Flow”.
The core problem that he discussed widely and for example made Adobe change a lot, was that when we calculate values while being gamma based, they create unreliable and weird results.
In a nut shell, or how I explained it in my hands on classes at the time, to simplify it: You have a human (gamma) level, and a Machine (linear) level. The human level needs anything translated to gamma, on the viewing devices, but internally, the computer have to “digest” anything in linear to get clean results.
More in detail, if you take the 256 values/channel of an 8bit file, they have a greater density in the middle field, where our attention is, and a lower density of values where the brighter parts are, naturally the amount in the darker areas is smaller anyway, for many reasons. As each value can go only into a small space of these 256 values, in other words, there is no 125.678 value, it is either 125 or 126, the precision after many operations will go down “south”. Now take anything besides the denser middle part and the jumps from one value to the other will be much larger, as they were stretched on the gamma curve. Banding would be one of those results.
Which leads at least to have 16 bit per channel, during production, even if 8bit/channel was provided and will be delivered, to avoid the 125 or 126 decision. The production pipeline must be larger (while talking about integer not float), to keep at least the given quality and not lower it.
So, why not store linear light values in an 8bit/channel integer file? Because the nature of linear would take half of the values for the brightest stop alone (stop as in photography), the next one a quarter, which means for anything below the two brightest stops, we would have 64 steps left. But it gets worse, as the next darker stop takes and 1/8th away, then the 1/16. So four stops of light would leave 1/16th for anything below. Not quite an impressive number. Whereby, the definition of the brightest and white, is just the limit of the sensor or an random decision of the artist. But I go already to far in details for the space I have to answer all questions. So:
This is all in a nut shell, and yes, I do not do gamma based work at all, as small or large production. The problems that might occur with 8bit/channel and the time to fix those, with dither or by applying noise even or other weird addition stuff, might take away the advantage of working in such limited space in the first place.
Colorspace is key because it mange all the color. This sounded weird, but the key here is, that any pixel value if color managed or not, will be color managed when it is, e.g., delivered to your screen for example. IF non is give, it will be assumed that the material was created by one. If you calibrate your monitor, so all the variables that a monitor has will be neutralized as far as possible, it creates a profile. What everyone wants, is, that what you can see on your screen is the same that someone else will see on his’ or her’s screen or projection.
Since we have different sizes of colorspace, Gamut, the definition of color for each RGB will vary based on it. IF the space is too small, values will be clipped and from there - there is no way back to retrieve it.
While working in a production pipeline, one does not want to have the small color space that the smallest delivery has, it should be larger than the best delivery has. So at no point we have a “125;126” effect, or again something like clipping.
The part of moving from one color space to the next is something that should be done with care and at least after have read the four options described in the Photoshop Manual about it.