@Medved Today, impressive name branded capacitors rated 4000 hours @135C temperature are available. You just won't meet them in cheapo lamps ))
Well that is the point, they are expensive. So you are then forced to cut corners elsewhere way more aggressively, so the reliability will suffer from there. Plus 4k hour is quite short for a light source these days, it means you have to restrict the capacitor core temperature to something below 100degC to get something usable (and you can not spend all the lifetime budget onjust one component, that would require the rest to have infinite MTBF in order for the whole lamp to meet the lifetime requirement). An there is another aspect: These are rather large components, so not that much suitable to "incendescent retrofit" format.
There are tricks to reduce the flicker when you want to avoid the electrolytic even without going boost SMPS. All revolves about a fact that the most apparent flicker is when you get sharp narrow pulses with long gaps, when the frequency is something you can not change.
Simplest circuit is to use lower LED drop and burn higher voltage on the regulator. But that is the lossiest one. But for low power it is acceptable (the extra heat is not that much), often it was the easiest way to use more expensive higher efficacy LEDs and stay with this ballast, the LEDs brought 2x improvement, while such ballast loses just about 20..30%, so you still gain while staying on budget.
Other option, appearing about 10 years ago, was to split the LEDs into sections and add switching transistors/diodes to dynamically either reconnect parallel/series, or just bypass a section when the instant mains voltage don't give enough room for the regulator to work. That way you get fluctuating light over the mains cycle, but when the gaps are not really dark, the flicker is not that obtrusive anymore.
A variant is, when you switch off the current when the voltage drop across the regulator becomes too high (e.g. with a two section LED string switched series/parallel the sinewave is in the part where the voltage is nearly double of a parallel mode, but not enough yet to let them operate in series). The drawback is way higher mains harmonic content, so could hit a regulation limit.
But generally these schemes vanished, as it require quite complex LED arrangement and very limited flexibility in that with a single ballast model.
Then there is the "good old buck SMPS". Simple, but needs special controller regulation scheme to manage both flicker, as well as the mains harmonics. But because it keeps the LED interface very simple, it becomes so u iversal by itself that it made sense to integrate those functions into dedicated ICs. None of the wiring and just very few components need to be modified to suit completely different spec LED string or optimization target. The whole circuit just needs few ceramic capacitors as HF filter, the mains rectifier, a fuse, this chip (maybe separate power transistor for higher power ballasts) and one resistor to program the average LED current, and it works with LED string from few 10'sV (when you preffer low flicker over losses) till some 200V (when you preffer low losses or high power with really small size).
Easiest to control circuit is a flyback SMPS (in DCM when you keep duty cycle constant ovsr time, you get nearly perfect power factor), but these need almost double voltage rated switches, which makes them not as attractive (voltage rating makes the transistors expensive,mainly when they need to be integrated on a monolithic die with the controller). Plus it needs bulkier coil, also an expense. But it can be inherently made to work with transformer instead of the simple inductor, so you get galvanic isolation of the LEDs from mains.
