[Rd] segfault with POSIXlt zone=NULL zone=""

[frederik at ofb.net]

Thanks for the reply, Duncan.

It looks like Martin can commit a patch faster than I can open a bug
report...

Frederick

On Tue, Dec 06, 2016 at 12:39:32PM -0500, Duncan Murdoch wrote:
> I agree this is a bug; R should never segfault.  I wouldn't call it a high
> priority one, since you can avoid the problem by not messing with R's
> internal structures.
> 
> It's unlikely to get fixed unless someone posts it as a bug report to
> bugs.r-project.org (because low priority bugs reported only on mailing lists
> get forgotten).
> 
> So please post a minimal example there, possibly accompanied with a patch.
> If you don't have an account, you can write to me privately and I'll set one
> up for you.  (We no longer allow people to create their own accounts because
> of abuse by spammers.)
> 
> Duncan Murdoch

On Tue, Dec 06, 2016 at 06:49:05PM +0100, Martin Maechler wrote:
> >>>>> Joshua Ulrich <josh.m.ulrich at gmail.com>
> >>>>>     on Tue, 6 Dec 2016 09:51:16 -0600 writes:
> 
>     > On Tue, Dec 6, 2016 at 6:37 AM,  <frederik at ofb.net> wrote:
>     >> Hi all,
>     >> 
>     >> I ran into a segfault while playing with dates.
>     >> 
>     >> $ R --no-init-file
>     >> ...
>     >> > library(lubridate); d=as.POSIXlt(floor_date(Sys.time(),"year")); d$zone=NULL; d$zone=""; d
>     >> 
>     > If you're asking about a bug in R, you should provide a *minimal*
>     > reproducible example (i.e. one without any package dependencies).
>     > This has nothing to do with lubridate, so you can reproduce the
>     > behavior with:
> 
>     > d <- as.POSIXlt(Sys.time())
>     > d$zone <- NULL
>     > d$zone <- ""
>     > d
> 
>     [..........]
>     
>     >> Hope I'm not doing something illegal...
>     >> 
>     > You are.  You're changing the internal structure of a POSIXlt object
>     > by re-ordering the list elements.  You should not expect a malformed
>     > POSIXlt object to behave as if it's correctly formed.  You can see
>     > it's malformed by comparing it's unclass()'d output.
> 
>     > d <- as.POSIXlt(Sys.time())
>     > unclass(d)  # valid POSIXlt object
>     > d$zone <- NULL
>     > d$zone <- ""
>     > unclass(d)  # your malformed POSIXlt object
> 
> Indeed, really illegal, i.e. "against the law" ... ;-)
> 
> Thank you, Joshua!
> 
> Still, if R segfaults without the user explicitly
> calling .Call(), .Internal()  or similar -- as here --
> we usually acknowledge there *is* a bug in R .. even if it is
> only triggered by a users "illegal" messing around.
> 
> an MRE for the above, where I really only re-order the "internal" list:
> 
> d <- as.POSIXlt("2016-12-06"); dz <- d$zone; d$zone <- NULL; d$zone <- dz; f <- format(d)
> 
> >  *** caught segfault ***
> > address 0x80000020, cause 'memory not mapped'
> 
> > Traceback:
> >  1: format.POSIXlt(d)
> >  2: format(d)
> 
> The current code is "optimized for speed" (not perfectly), and
> a patch should hopefully address the C code.
> 
> Note that a smaller MRE -- which does *not* re-order, but just
> invalidate the time zone is
> 
>   d <- as.POSIXlt("2016-12-06"); d$zone <- 1; f <- format(d)
> 
> ------
> 
> I have now committed a "minimal" patch (to the C code) which for
> the above two cases gives a sensible error rather than a
> seg.fault :
> 
>   > d <- as.POSIXlt("2016-12-06"); d$zone <- 1 ; f <- format(d)
>   Error in format.POSIXlt(d) : 
>     invalid 'zone' component in "POSIXlt" structure
> 
>   > d <- as.POSIXlt("2016-12-06"); dz <- d$zone; d$zone <- NULL; d$zone <- dz; f <- format(d)
>   Error in format.POSIXlt(d) : 
>     invalid 'zone' component in "POSIXlt" structure
>   > 
> 
> I guess that it should still be possible to produce a segfault
> with invalid 'POSIXlt' structures though.
> 
> Martin
>