On 2016-06-28 00:10, Greg Stark wrote:
On Mon, Jun 27, 2016 at 11:02 PM, Tom Ivar Helbekkmo <tih%hamartun.priv.no@localhost> wrote:Well the awk core dumps were presumably [...]I don't know if this is still an issue, but there's been a problem in the past, specific to the VAX because of its different floating point implementation, where if awk sees a string token beginning with the characters "nan", case independent, it'll read it as a NaN float, and promptly dump core.Well there's two problems with the theory that it's floating-point related. 1) It's non-deterministic. I would expect most circumstances like you describe to be repeatable.
The problem was happening when building netbsd, and I used to just rerun the build process with the update options, and each time I got a bit further. Which is why I asked if you did a clean in between, since there are definitely situations where a rebuild will not give the same error.
and 2) things like 1/0 and 0/0 on VAX don't cause SIGILL, they cause floating point exceptions which would indeed core-dump but I'm seeing SIGILL. It happens that log(0) and other libm functions do indeed signal errors via SIGILL but it seems somewhat less likely that awk would be doing anything like that to generate an Inf or NaN.
Believe it or not, but awk was doing exactly that - a log(0), which would cause a silent NaN on other machines.
Johnny
--
Johnny Billquist || "I'm on a bus
|| on a psychedelic trip
email: bqt%softjar.se@localhost || Reading murder books
pdp is alive! || tryin' to stay hip" - B. Idol