Re: [Boost-users] Boost 1_59_0_b1_rc1 regression in spirit::qi

On 7/15/15 3:51 AM, Lee Clagett wrote:

On Tue, Jul 14, 2015 at 8:10 AM, Oswin Krause mailto:Oswin.Krause@ruhr-uni-bochum.de> wrote:

Hi,

On 2015-07-14 05:54, Marshall Clow wrote:

Release candidate files for 1.59.0 beta 1 are available at http://boost.cowic.de/rc/ [1]

As always, the release managers would appreciate it if you download the candidate of your choice and give building it a try. Please report both success and failure, and anything else that is noteworthy.

Boost compiles fine, OS Arch Linux, x86_64.

However, i got a regression in my software using boost::spirit::qi

See attached the program (i guess it does not qualify minimal, but i do not have time to trace it further), output of g++/valgrind for boost 1.58 and for 1.59. There seems to be a signed/unsigned comparison issue and valgrind detects a conditional jump that depends on unintialized memory. In my code this showed up as 4 values read from the parsed file have wrong values albeit the 200 or so other values before and after are fine.

Best, Oswin

I found the issue, and issued a pull request to fix it. Joel let me know if you want a test case; a variable currently unitialized (presumably intentionally) would have to be value-initialized to make it reliably catch this particular case (and then it could suppress errors if the float is `0`).

Thanks for the fix, Lee. Yes, we need a test case, please. I'll merge the PR now but please add the test case into Qi's tests. Thanks for spotting this and reporting the problem, Oswin. This is indeed a showstopper. Marshall, how shall we deal with this? Shall I merge the fix immediately to master? regards, -- Joel de Guzman http://www.ciere.com http://boost-spirit.com http://www.cycfi.com/