R devel - Jun 2004 - Bug in parse(text = <long polynom>) (PR#7022)

Merci beaucoup, Jean,
for the bug report -- which I'm no "completeing" to R-bugs
>>>>> "Jean" == Jean Coursol
<coursol@cristal.math.u-psud.fr>
>>>>>     on Thu, 24 Jun 2004 15:22:37 +0200 (CEST) writes:
    Jean> I was exploring the polynom library with students:

 <and found a segmentation fault from parsing a long expression>

The following is reproducible also with the current version of R
1.9.1 [on RHEL Linux]

horner <- function(p) {
        a <- as.character(rev(unclass(p)))
        h <- a[1]
        while (length(a <- a[-1]) > 0) {
            h <- paste("x*(", h, ")", sep = "")
            if (a[1] != 0)
                h <- paste(a[1], " + ", h, sep = "")
        }
        h
}

library(polynom)

x <- polynomial()
z <- (1+x)^100
zh <- horner(z)

nchar(zh)
## [1] 2404

parse(text = zh) # => Segmentation fault

## where Jean wrote  '(it ran one time !!!)'
## and it happens the first time for me.

maechler@stat.math.ethz.ch writes:
> Merci beaucoup, Jean,
> for the bug report -- which I'm no "completeing" to R-bugs
But you're still requiring library(polynom) for triggering the bug. If
we are to be sure that it is not a bug in that package but a bug in R,
you need to include the definition of at least polynomial() with the
instructions to reproduce the effect...

[snip]
> library(polynom)
> 
> x <- polynomial()
> z <- (1+x)^100
> zh <- horner(z)

-- 
   O__  ---- Peter Dalgaard             Blegdamsvej 3  
  c/ /'_ --- Dept. of Biostatistics     2200 Cph. N   
 (*) \(*) -- University of Copenhagen   Denmark      Ph: (+45) 35327918
~~~~~~~~~~ - (p.dalgaard@biostat.ku.dk)             FAX: (+45) 35327907

On 25 Jun 2004, Peter Dalgaard wrote:
> maechler@stat.math.ethz.ch writes:
> 
> > Merci beaucoup, Jean,
> > for the bug report -- which I'm no "completeing" to
R-bugs
> 
> But you're still requiring library(polynom) for triggering the bug. If
> we are to be sure that it is not a bug in that package 
There seems to be in that it is not using the R code from `S Programming' 
p.95, according to the original report.
> but a bug in R,
> you need to include the definition of at least polynomial() with the
> instructions to reproduce the effect...
> 
> [snip]
> > library(polynom)
> > 
> > x <- polynomial()
> > z <- (1+x)^100
> > zh <- horner(z)
Hmm, horner is not in the package!   It is a function internal to 
as.function.polynomial.

as.function(z) segfaults at

#0  R_TextBufferGetc (txtb=0xffda00)
    at /users/ripley/R/cvs/R-devel/src/main/iosupport.c:232
#1  0x080c7623 in text_getc () at ./gram.y:1011
#2  0x080c6eed in xxgetc () at ./gram.y:291
#3  0x080c7bc5 in token () at ./gram.y:1496
#4  0x080c85c5 in Rf_yylex () at ./gram.y:1894
#5  0x080c8bef in Rf_yyparse () at /usr/share/bison/bison.simple:573
#6  0x080c997e in R_Parse1 (status=0xbfffda58) at ./gram.y:941
#7  0x080c9aad in R_Parse (n=-1, status=0xbfffda58) at ./gram.y:1076
#8  0x080c9c1e in R_ParseVector (text=0xffffffff, n=-1, status=0xffffffff)
    at ./gram.y:1153
#9  0x0813c3d9 in do_parse (call=0x8500c00, op=0xffffffff, args=0x84fb898,
    env=0x8ed385c) at /users/ripley/R/cvs/R-devel/src/main/source.c:68

so that's definitely a bug in R. It is being asked to parse a very long
line and a highly-nested expression.  I suspect the latter is the problem.

-- 
Brian D. Ripley,                  ripley@stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595

>>>>> "PD" == Peter Dalgaard
<p.dalgaard@biostat.ku.dk>
>>>>>     on 25 Jun 2004 09:32:48 +0200 writes:
    PD> maechler@stat.math.ethz.ch writes:
    >> Merci beaucoup, Jean, for the bug report -- which I'm no
    >> "completeing" to R-bugs

    PD> But you're still requiring library(polynom) for
    PD> triggering the bug. If we are to be sure that it is not
    PD> a bug in that package but a bug in R, you need to
    PD> include the definition of at least polynomial() with the
    PD> instructions to reproduce the effect...

well, try this (without package polynom):
> parse(text="1 + x*(100 + x*(4950 + x*(161700 + x*(3921225 +
x*(75287520 + x*(1192052400 + x*(16007560800 + x*(186087894300 +
x*(1902231808400 + x*(17310309456440 + x*(141629804643600 + x*(1050421051106700
+ x*(7110542499799200 + x*(44186942677323608 + x*(253338471349988672 +
x*(1345860629046814720 + x*(6650134872937201664 + x*(30664510802988208128 +
x*(1.32341572939212e+20 + x*(5.3598337040381e+20 + x*(2.04184141106213e+21 +
x*(7.33206688517765e+21 + x*(2.48652703062547e+22 + x*(7.97760755659004e+22 +
x*(2.42519269720337e+23 + x*(6.99574816500972e+23 + x*(1.91735320078044e+24 +
x*(4.99881370203473e+24 + x*(1.24108478119483e+25 + x*(2.93723398216109e+25 +
x*(6.63246383068634e+25 + x*(1.43012501349174e+26 + x*(2.94692427022541e+26 +
x*(5.8071742972089e+26 + x*(1.09506715318796e+27 + x*(1.97720458214493e+27 +
x*(3.42002954749394e+27 + x*(5.67004898663469e+27 + x*(9.01392403003463e+27 +
x*(1.37462341458028e+28 + x*(2.01164402133700e+28 + x*(2.82588088711626e+28 +
x*(3.81165328959867e+28 + x*(4.93782357970737e+28 + x*(6.14484712141362e+28 +
x*(7.3470998190815e+28 + x*(8.4413487283064e+28 + x*(9.320655887505e+28 +
x*(9.8913082887808e+28 + x*(1.00891344545564e+29 + x*(9.8913082887808e+28 +
x*(9.320655887505e+28 + x*(8.4413487283064e+28 + x*(7.3470998190815e+28 +
x*(6.14484712141362e+28 + x*(4.93782357970737e+28 + x*(3.81165328959867e+28 +
x*(2.82588088711626e+28 + x*(2.01164402133700e+28 + x*(1.37462341458028e+28 +
x*(9.01392403003463e+27 + x*(5.67004898663469e+27 + x*(3.42002954749394e+27 +
x*(1.97720458214493e+27 + x*(1.09506715318796e+27 + x*(5.8071742972089e+26 +
x*(2.94692427022541e+26 + x*(1.43012501349174e+26 + x*(6.63246383068634e+25 +
x*(2.93723398216109e+25 + x*(1.24108478119483e+25 + x*(4.99881370203473e+24 +
x*(1.91735320078044e+24 + x*(6.99574816500972e+23 + x*(2.42519269720337e+23 +
x*(7.97760755659004e+22 + x*(2.48652703062547e+22 + x*(7.33206688517765e+21 +
x*(2.04184141106213e+21 + x*(5.3598337040381e+20 + x*(1.32341572939212e+20 +
x*(30664510802988208128 + x*(6650134872937201664 + x*(1345860629046814720 +
x*(253338471349988672 + x*(44186942677323608 + x*(7110542499799200 +
x*(1050421051106700 + x*(141629804643600 + x*(17310309456440 + x*(1902231808400
+ x*(186087894300 + x*(16007560800 + x*(1192052400 + x*(75287520 + x*(3921225 +
x*(161700 + x*(4950 + x*(100 +
x*(1))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))")

>>>>> "MM" == Martin Maechler
<maechler@stat.math.ethz.ch>
>>>>>     on Fri, 25 Jun 2004 10:19:19 +0200 writes:
>>>>> "PD" == Peter Dalgaard
<p.dalgaard@biostat.ku.dk>
>>>>>     on 25 Jun 2004 09:32:48 +0200 writes:
    PD> But you're still requiring library(polynom) for
    PD> triggering the bug. If we are to be sure that it is not
    PD> a bug in that package but a bug in R, you need to
    PD> include the definition of at least polynomial() with the
    PD> instructions to reproduce the effect...

    MM> well, try this (without package polynom):

    >> parse(text="1 + x*(100 + x*(4950 + x*(161700 + x*(3921225 
....

and then I had the full length 2404 string which---I had
"known in advance" (Murphy's law)--- has been thrown up
before it got back to R-devel, actually by my mailer already.

Now this should be easier to pass through:

 tt <- tempfile() 
 download.file("ftp://ftp.stat.math.ethz.ch/U/maechler/R/longpoly.rda",
tt)
 load(tt)
 zh # look at the long "polygon string"

 parse(text=zh) ## seg.fault

--
Martin