On Mon, Jul 29, 2013 at 05:46:35PM -0700, L. David Baron wrote:
> I think the important characteristic here that we need is that we
> need to test builds compiled with -DDEBUG.

Yes.

> It doesn't matter if the builds are optimized or not (given that I'm
> under the impression that running non-optimized B2G is unacceptably
> slow)

Agreed; there's no value in unoptimized builds here.  Also, debug+noopt
with the regular Android ICS toolchain is known to fail by exhausting
the address space of the linker.

> and I'd think we don't want debug symbols to be part of the binaries
> on the device (which is a characteristic of debug builds on desktop,
> though I'd hope that the build system already does something
> reasonable here).

We're already building full debug information by default and stripping
the binaries shipped to the device; this is normal for embedded and
kernel development, and reasonably well supported by the usual tools.

> I wrote a blog post explaining more generally why I think this is
> important:  http://dbaron.org/log/20130729-debug-builds

One thing which especially concerns me is the lack of consistency within
the project -- we have shared infrastructure that's written with the
assumption that its MOZ_ASSERTs and JS_ASSERTs are true under conditions
covered by the test suite (let alone by merely starting Gecko) -- and an
entire class of platforms (not just b2g, but all ARM targets) where this
is not the case.

Also, there are two related issues here: continuous automated testing of
debug builds, and developers using debug builds for their regular work
(other than performance work).  Having the former without the latter
seems, at the risk of stating the obvious, as if it will result in
general frustration.

> Are there plans to have B2G automated tests running on debug builds?
> What's the status, and what's in the way of it happening?

Test infrastructure issues were previously mentioned.  An uncertain but
nonzero number of things are broken on debug builds because they're
not well tested yet.  And, as mentioned, performance issues allegedly
sufficient to discourage using debug builds for general development
work.

Bug 845886 exists, to change the default developer build; I took the
bug, but at this point I don't expect to have time for it for at least
the next 2 weeks, so if someone else can get to it sooner, feel free to
reassign.

Also, when I brought this up on #developers a few weeks ago, I was told
that there are known issues with detecting assertion failures on b2g
(and maybe Android as well), but that may have been about a different
kind of "assertion" than the crash-on-failure ones I had in mind.

--Jed

_______________________________________________
dev-b2g mailing list
[email protected]
https://lists.mozilla.org/listinfo/dev-b2g

Reply via email to