[PATCH v1] RISC-V: Support XVentanaCondOps extension

lkcl luke.leighton@gmail.com
Wed Apr 20 17:55:01 GMT 2022


sorry this is out of replyto because i am subscribed digest.

> Subject:	Re: [PATCH v1] RISC-V: Support XVentanaCondOps extension
> Hi Philipp:

> I believe we have a consensus among most GNU toolchain maintainers for
> accepting vendor extension to upstream / master branch,

my understanding of custom extensions is that they are intended never, under any circumstances, to hit "upstream", due to the risk of creating massive conflict and confusion due to dominance of one extension over another, particularly if that custom extension achieves extremely commonly-used and high profile public status.

[a dynamic plugin system on the other hand is a completely different matter]

i warned the RISC-V Foundation about this scenario.  they ignored my warnings.  two years later as a recent article online outlines, the fragmentation has already occurred because this 3rd category (abuse of custom extension opcode space for *high profile* common public usage) has, as i warned would happen, in fact occurred.

custom extensions were *supposed* to be hard-forks of the entire toolchain, maintained by a proprietary vendor, at their cost and expense, for their benefit and their benefit only.

[a dynamic plugin system helps such proprietary vendors to reduce costs of maintaining such private hard forks]

if you accept proprietary (rogue, unauthorized) custom plugins into the toolchain you risk destruction of the entire ecosystem for everyone.

[by complete contrast a dynamic plugin system helps keep such rogue extensions "at bay", crucially *without* giving the false and misleading impression that the extensions have been "authorised" through upstream acceptance]

if unfamiliar with this scenario recall the nightmare opcode conflict of altivec for powerpc, of 20 years ago.  ask around.

l.


More information about the Binutils mailing list