Re: In a cross-compile do host and build need exactly the same tools?

[Joseph Myers <joseph at codesourcery dot com>]

On Mon, 22 Oct 2018, Carlos O'Donell wrote:

> It looks like we would require python 3.4 on the host, even
> in a cross-build environment to get clean test results.
> 
> Is it a general principle that we require exactly the same tools on build
> and host to get clean test results?

It's known that if the host doesn't have the Python program found by 
configure on the build system then the pretty-printers tests fail.  
Likewise one rpcgen test using cpp.  See 
<https://sourceware.org/glibc/wiki/Release/2.28#Architecture-independent>.

(However, tests run mtrace on the build system, so it doesn't matter for 
test purposes whether the host has Perl at the same path it was found on 
the build system - although, logically, it's a defect to share 
configuration between paths to Perl on the two systems, and if we change 
mtrace into a Python 3 script it would similarly be a defect to share 
configuration of the Python 3 path; using standard paths 
"#!/usr/bin/<interpreter>" or "#!/usr/bin/env <interpreter>" or else 
having a way to configure such paths separately for the two systems would 
be best.)

-- 
Joseph S. Myers
joseph@codesourcery.com