Re: Upstreaming glibc for OpenRISC

["Joseph S. Myers" <joseph at codesourcery dot com>]

On Thu, 17 Jul 2014, Christian Svensson wrote:

> Due to the nature of the CPU only being available on FPGAs and in
> simulators I have not been able to run the test suite (too slow/too
> little RAM to compile natively), which has been one of the things
> putting me off from reaching out.

glibc supports cross testing over SSH.  You may need to set a large 
TIMEOUTFACTOR (my experience suggests TIMEOUTFACTOR=50 is sufficient for 
FPGA-based MIPS processors, for example), and there are a handful of tests 
that fail if you have under about 0.5GB RAM (you may have more problems if 
there is too little memory to run localedef on the OpenRISC system, as 
failure to generate locales used in testing will cause tests using those 
locales to fail as well, though you could work around that by generating 
the locales using localedef - from the same glibc version and with the 
right endianness option - on the build system first).

> What I'm asking is: at what stage do you want to accept the code for inclusion?

The testsuite should of course have been run and any failures understood 
(and fixed if they indicate port-specific bugs).  The port-specific files 
should be up-to-date with all global changes made since you started the 
port.  The port should follow recommended good practice for ports, 
including setting the minimum symbol version to whatever glibc version it 
goes in at.

-- 
Joseph S. Myers
joseph@codesourcery.com