[R] .Internal(La_rs(x, FALSE)) crashes R after long (reproducible) script on windows only

[Anthony Damico]

hi, thank you,

at the point that the corruption exists, the line
`.Internal(La_rs(x,FALSE))` actually breaks without needing `eigen`

i have provided a reproducible example but agree it might not be minimal --
i did try removing various sections, each time the bug unfortunately
vanished.  note the february 22nd edit: even interspersing the script with
the line that triggers the crash prevents the crash in the first place!

i think this occurs in C and not R, and would appreciate pointers about how
one might do that?  the only advice i've have is rebuilding R with a debug
build and gdb, but this seems like a huge lift --  are there any shortcuts
here for someone mostly unfamiliar with C code and even setup?  general
advice on this thread might also help me crack this case  :)

stackoverflow.com/questions/35455135/general-suggestions-on-debugging-internal-in-r

thanks for your time

On Tue, Feb 23, 2016 at 8:22 AM, Duncan Murdoch <murdoch.duncan at gmail.com>
wrote:

> On 23/02/2016 7:49 AM, Anthony Damico wrote:
>
>> hi, does anybody have a clue why .Internal(La_rs(x,FALSE)) is getting
>> corrupted (actual detonation occurs within La_solve_cmplx within Lapack.c)
>> on windows but not mac/unix?
>>
>> i have provided two (long) scripts that reproduce the problem and a third
>> script modified to trigger the crash that unfortunately does not reproduce
>> the problem
>>
>>
>> http://stackoverflow.com/questions/35447971/internalla-rsx-false-crashes-r-after-long-reproducible-script-on-windows
>>
>
> Just two comments:
>
>  - Your post suggests you're calling .Internal() yourself, but that's not
> the case.  So your question should be about why eigen() crashes R.
>
>  - If you need a long script to trigger the error, I'd assume there's
> something wrong in that script. Your script uses several contributed
> packages, so the problem could be there. Shorten it to a minimal
> reproducible example that doesn't use any contributed packages.  If you
> can't leave out the packages, try to reduce it to just one, and ask the
> maintainer of that package about it.
>
> Duncan Murdoch
>

	[[alternative HTML version deleted]]