RFC V4 test-in-container

[DJ Delorie]

Joseph Myers <joseph@codesourcery.com> writes:
> I remain concerned about copying /bin/sh rather than having a local 
> sh-substitute.  Can using the newly built dynamic linker to trace what's 
> required by /bin/sh really work even in as simple a case as testing i386 
> glibc on an x86_64 system, where /bin/sh is a 64-bit binary?

I can write a /bin/sh substitute as long as none of our tests test *it*
and not just the calls that require it ;-)

I think one caveat is that we need to decide where and how to document
it so that future test writers don't assume full /bin/sh functionality.

But I suspect it would work; the scriptlet copies libc.so from the host
system as well as everything else.  If it's the "wrong" libc.so (/lib vs
/lib64) we won't override it with our own.  It would only be a
"sufficiently reliable coincidence" though.