[Toybox] LLVM sanitizers

[Rob Landley]

On 07/21/2016 12:33 AM, Andy Chu wrote:
> expr is in pending, but ships on Android.

Yes, that is a significant issue I need to address. However, "issue"
isn't necessarily "the real problem". The _problem_ is I have many
things to do and a finite amount of time/energy to do them, but I'm
working on it.

Right now I'm trying to get grep to where android can use it, and get dd
cleaned up and promoted out of pending. After that expr is near the top
of the list.

When you say "a thing in pending is broken" I tend to go "yes, that's
why it's in pending". When you say "prioritize this thing in pending"
what I prioritize is getting it out of pending.

> I hope that was clear!  I thought you were just busy, but I think you
> never really understood what I was saying and what the tools do.

I want to triage the testing directory, going through the existing tests
command by command and making sure that each test is a good test for the
command. (Are we passing the tests we have? Can TEST_HOST reasonably be
applied? Are we testing everything the command does? Are we testing
interesting failure paths?)

My major todo item is getting tests requiring root access and/or a known
host environment implemented and working under aboriginal linux images.
(For "top" I may need a snapshotted /proc directory in a tarball, and
/proc symlinks switching from directory to directory with somesort of
synchronization mechanism. I probably have to create a pty, run top with
an essentially infinite timeout, and feed a space through the pty to
force a rescan, and also make sure the code is taking its timing info
from the proc data not from calling gettimeofday() or similar so the
test scriptisn't perturbing the output.)

That can of worms is the kind of thing I'm worrying about when I look at
the test suite. Can I make dummy interfaces in a qemu instance that
ifconfig can attach ipv6 addresses to, and if so can I insulate it from
kernel version skew if I upgrade the kernel version the test is running
under?)

You instead want to do a major redesign of the testing directory based
on this week's coverity/valgrind variant, which is a different area
entirely and not something I personally ever expect to use. Your
redesign would put me further from _my_ goals, adding significant
complexity I wouldn't use and thus wouldn't regression test and thus
almost certainly would break if I changed anything. (Not just changes in
the testing directory, you add significant complexity to a makefile
that's otherwise mostly just a wrapper around shell scripts in
scripts/*.sh to provide an expected user interface, the way "make
test_sed" is just a wrapper around "scripts/test.sh sed". With your
patch, the makefile seems to be the ONLY way to call some of the new
functionality, and/or your wrappers call the makefile instead of the
other way around.)

> If there were any parts of my messages that weren't clear, I'm happy to
> clarify.

I lost about 4 weeks in the past 2 months to $DAYJOB melting, and it's
entirely possible I'll be spending 6 months in San Diego working an
unrelated contract until they get their financial house in order.

More recently the Con Crud I picked up at Texas Linuxfest seems to have
morphed into tonsilitis (which is disconcertingly like strangling except
my airway isn't blocked, and I keep wanting a 6 hour nap in the middle
of the day).

I'm working as fast as I can, but I don't necessarily have the same
priorities you do.

> Andy

Rob