There's nothing inherently non-RISC there.
Indirect addressing was evaluated for RISC-V, and found to not be worth the required added complexity and encoding space.
Whereas RISC-V actually adopted variable-length instruction size, because it was found to be highly beneficial to code density, and doable in a manner that does not add too much decode complexity, in neither small nor large micro-architectures.
wtf?
If it was ever suggested by someone (and would NEVER be by Krste, Andrew, Yunsup, or Dave) then any evaluation would take less than 1 second before the "HELL NO!"
> Whereas RISC-V actually adopted variable-length instruction size
TWO instructions sizes, in the currently ratified ISA. And you can tell the size by looking at just 2 bits in the first byte of the instruction (the LSBs).
That is very far from "variable-length" in the sense of VAX or M68k or x86.
> ... was found to be highly beneficial to code density, and doable in a manner that does not add too much decode complexity
Commercially known in microprocessors since at least SuperH 2A and Arm Thumb2. Not to mention 3 lengths (2, 4 or 6 bytes) in IBM 360, also decodable from just 2 bits in the instruction. Also CDC 6600 (arguably the first RISC), Cray-1 (also recognizably RISC), the first version of IBM 801, and Berkeley RISC-II (obviously familar to the RISC-V designers).
And yes, RISC-V designers believe that addressing more complex than base+offset is a net loss in processor efficiency. You might need very slightly fewer instructions overall in a program, but it costs more transistors, silicon, and therefore dollars, and might even lower the achievable clock speed by more than the saved instructions.
Some companies implementing RISC-V don't quite believe that (e.g. Andes, THead) and add more complex addressing modes as custom extensions. Once they've spent the silicon then you might as well use it -- same with WCH's shadow registers or automatic push/pop for interrupt handling (depending on the core) -- but it's not proven it's actually the best use of additional silicon [1].
Let the market decide.
If complex addressing wins then RISC-V always has the possibility to add a standard extension. It's much harder to take unneeded features away.
[1] admittedly the equation is different with a single-core microcontroller in a package, vs a chip with large numbers of cores (whether all the same, or all RISC-V, or not). On a stand-alone microcontroller the die size is often determined by fitting the pads around the outside, and if you don't use all the silicon inside that then it's wasted.