The issue is in the way a character string is initialized from the first bytes of memory on the stack. In the past, the first byte was predictably a non-integer. With libstdc++-3.2.3-49 and gcc-c++-3.2.3-49, the first byte is predictably "4". We found this because we passed it directly to ATOI() which understands integers as valid input and non-integers as junk. In this case, the fault was on our end for disobeying the docs and assuming that junk would be junk. However, the implications are interesting if other, more high profile programs, make that same assumption. Also, I didn't want to have called you with a "bug" without giving you some kind of followup.
From the incomplete description this sounds like you are triggering undefined behaviour, so anything can happen. That would mean NOTABUG resolution. If you post a (selfcontained) testcase what exactly you're talking about, I'll look at it, but without a testcase this will be NOTABUGed soon.
Assuming NOTABUG, if you have self-contained testcase, please reopen.