Comment 1:

Labels changed: added size-xl.

Comment 2:

Another reason to do that is that cacheflush calls during closure generation are
incredibly expensive, for example on linux/arm. See CL7062045 for an example where
removing closures from GcImporter speeds up package typechecking by 2x.

Comment 3:

fixing this will also reduce security surface for cgo binaries because the elimination
of executable heap.
besides, runtime.cacheflush has tripped me twice.
we just need to discuss the interaction between this and method functions:
should (T).M return a wrapper function (takes an extra closure data
parameter) or not?

Comment 4:

Are closures the only place where we use runtime.cacheflush ? If we
didn't have to support this runtime facility any more that would be a
great boon to people porting Go to new platforms. For example, how
does one do a cacheflush on plan9 ? On Solaris ?

Comment 5:

Re #4, yes, closure creation is the only place we use runtime.cacheflush.

Comment 9:

This issue was closed by revision b1b67a3.

Status changed to Fixed.