glslang preprocessor pragmas

the glslang spec says that pragmas are implementation dependent. This can reduce portability of code.
Can we somehow establish a name mangling convention, like that used for extensions?
For example,
#pragma Rock(on)
might mean entirely different things for
ATI and NVIDIA compilers.
Instead, they would be
#pragma NV_Rock(on)
#pragma ATI_Rock(on)

Should this be called out in the spec?