On 05/03/2017 12:47 PM, Martin Sebor wrote:
I can do some of it but not all of it. The work doesn't involve
just building the compiler but also running the tests and fixing
up regressions in those that are written to expect the unquoted
diagnostics. I don't have the ability to run the test suite on
all the targets, or the bandwidth to run it on a subset of them
beyond powerpc64 and x86_64. So I'm hoping to find a way for
the core of the patch to be checked in and for the cleanup to
proceed subsequently, as target maintainers find time.
Just a note. It's not terrible to run non-execute tests with cross
tools. In fact, it's fairly easy. I'm happy to walk you through the
process.
Long term I think we want to move away from config-list.mk and instead
use a buildbot. I've been experimenting with jenkins. I've got a
series of jobs that will build the *-elf, *-rtems, *-gnu and *-linux
targets. binutils, then gcc, then newlib/glibc.
The newlib (*-elf and *-rtems) pipeline is pretty solid at this point
and covers ~70 targets from config-list.mk building through newlib.
The glibc pipeline is more problematic, but improving.
It's about ready to leave my little cluster of slaves and relocate to
the servers in Toronto.
It should be fairly straightforward to take that jenkins work and tie in
testsuite runs. Even if we just start with compile.exp and expand one
day (via the simulators) to the execution tests.
Jeff
