On Thu, Mar 24, 2016 at 4:48 AM, Nick Wellnhofer <[email protected]> wrote:

> [ ] +1 Release RC 2 as Apache Clownfish 0.5.0.
> [ ] +0
> [ ] -1 Do not release RC 2 as Apache Clownfish 0.5.0 because...

-0, withholding approval until an issue with Str_To_F64 is resolved, see
below.

Sums and sigs OK.

Tests pass on OS X and Amazon Linux with Lucy's test_all.sh.

I confirmed that the release tarball matches the RC git tag.

I inspected the diff between RC1 and RC2 and everything looks good.  The
copyright date has been updated.

The vast majority of the compiler warnings are gone -- yay!  There are still
some unused var warnings under the Perl bindings, which are addressed on
master.  Not a big deal.

The `test_valgrind` build target passes for the Perl bindings of both CFC and
the Clownfish runtime.  However...

A problem with a Str_To_F64 has been revealed by Lucy's `test_valgrind`
target.  There's a call to `strtod` in Clownfish's Str_To_F64 reads past the
end of the buffer consistently now that the nul-termination is gone from
Clownfish's String implementation.  See here for the fix on master:
https://github.com/apache/lucy-clownfish/commit/4dab61a57

This wasn't a critical problem in 0.4.x and earlier because while it was
possible to read beyond the end of the String's logical content in esoteric
cases (such as a substring), even then the actual C buffer was always
nul-terminated and `strtod` would always stop.  What's different in 0.5.x is
that `strtod` can actually read into invalid memory.

I think we may want to cherry-pick that commit and put up a new RC.  Sorry for
not flagging this earlier.

Marvin Humphrey

Reply via email to