Re: RFC: Test optimizations for multiple architectures

[David Miller <davem at davemloft dot net>]

From: "H.J. Lu" <hjl.tools@gmail.com>
Date: Thu, 11 Nov 2010 15:50:29 -0800

> 1. Roland suggestion is orthogonal to mine.
> 2. Roland suggestion may not be easy to implement safely.
> 3. Even if it works, use it in a test script to cover all cases is complex.
> 4. We can enable --enable-test-multi-arch by default and change
> "make install" to remove libc-test.o from libc.a and libc-test.os from
> libc.so when installing libc.
> 
> Again, this only covers the testing part, not the granting user control
> part, which is a separate issue.

I think you're being unreasonable and ignoring some fundamental
aspects of these ifunc routines.

By very definition, you must know which cpu mask feature bits
are appropriate for the ifunc routine you are writing.

Therefore you know exactly which cases need to be added to the
test scripts.

You also underestimate the true cost of doing all of the ifunc
case builds for a distribution maintainer, it is an excessive
and unnecessary burdon to place upon them.

So as I see it your argument is null and void.