[Cc to -support]

I think I've finished testing how to build rustc-1.19.0 (including
cargo).  This has all been done with DESTDIR installs, but the
results look ok and so I plan to do the following on my next build
(unless there are strong reasons not to), and to then put this in
the 8.1 book (provided firefox-55 builds and works ;-)

Instead of building llvm-3 in /opt (current rustc supports llvm-4,
but as soon as llvm-5 is out that will probably not be usable, at
least until 1.20 or maybe 1.21), with shared libs, use the shipped
version of LLVM in rust (no, I didn't look to try to see which
version that is), with its static libs which of course makes things
bigger.  And instead of using configure, put all the necessary
non-default options into config.toml.  That includes installing
cargo from the rustc build.  One of the options limits the target to
X86 (so, no use if you want to build for Aarch64, Mips, Powerpc,
SystemZ (IBM mainframes) or some other targets), but that is similar
to how the current build has been limited (/opt/llvm3 only builds
for x86_64, or for i686 if anybody has used it for that).

This change has one major benefit : the tests work (always a good
sign when bootstrapping a compiler).  With a gcc-7.1 system and LLVM
from /opt/llvm3/bin I had one test which burned CPU and looked as if
it would run forever, one failure because that didn't support a
thumbv6 target, and two other failures which looked slightly
worrying (no idea what wa wrong) - that was after I identified and
deleted the test that burned CPU.

On 8.1-rc2 the test results from a configure / opt/llvm3 build were
much worse, it seemed to be burning CPU in more than one test, and I
never managed to identify which tests were involved (it only reports
the test AFTER it decides if it passed or failed, and killing the
currnet rustc process (which was how I had identified the problem
test yeaterday) stopped the tests dead.  The builds with the shipped
LLVM both passed their testsuites.

In rustc-1.16.0, of ocurse, the selection of which tests were to be
run appeared to be random - I think they've fixed that, the log from
my completed testsuites all show a total of 14029 tests to be run
(or 14028 when I removed that CPU-burning test).  Similarly, the gdb
tests failed without gdb, but some still failed even when it was
present.  I thought I didn't have gdb on yesterday's builds, but the
system was old, maybe I had installed it and forgotten - if its
omission does lead to failures, I'll find out on my next system ;)

Further details in
http://wiki.linuxfromscratch.org/blfs/ticket/9652, please speak up
if you do not think this is a good way to proceed.

ĸen
-- 
Truth, in front of her huge walk-in wardrobe, selected black leather
boots with stiletto heels for such a barefaced truth.
                                     - Unseen Academicals
-- 
http://lists.linuxfromscratch.org/listinfo/blfs-support
FAQ: http://www.linuxfromscratch.org/blfs/faq.html
Unsubscribe: See the above information page

Reply via email to