glm crash (PR#197)

[Cor.Berrevoets@gironet.nl]

Dear bug-busters,

This week I've posted a message to the R-help mailing to inquire if
anybody had experienced alike me. When I run the following GLM:
glm(n~s+c+m, family=quasi(link=log, variance=mu))
several times using exactly the same dataset R crashes at random. I have
tried to probe into the problem by tracking the response times of the
above function with system.time (e.g. system.time(glm ....))

I noticed that the response times increased in subsequent calls to the
function, at some (random) point leading to a crash. So it looked as if
R was "leaking" somewhere. Then I started all-over again (restarted PC)
and released before the glm-command a gc-command to collect the garbage.
That works rather well and the crashing seems less frequent (mostly
after 3-4 times issueing the glm-command) . Also the increasing response
times changed into a stable component with one big exception, the First
Call. Every time after fully re-starting R I issued gc() and then the
glm-command . This call is about 1/3 or so of the time spend in the
subsequent calls. Looking at the Ncells/Vcells hardly any difference
exists, on average the Ncells/Vcells (virgin state of the program) are
about 50 cells bigger than after the subsequent glm-commands.  In the
table below you can see the effects I just mentioned, notice the "odd"
behaviour of the default Vsize.

Table: response times glm function at varying Vsizes.

Vsize    1st response    subsequent
default            47.57        27.96
8M                39.27        62.15
16M              38.44        62.37
32M              13.34        32.77
64M              10.13        34.28
128M            11.45        34.21

Im using R 0.64.1 on Win95..

Maybe someone of your team can tell me more on the nature of this
"thing".

Hoping not to bug to much,

greetings

Cor Berrevoets





-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-devel mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To: r-devel-request@stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._