I don't know about the technicalities, but Peter Dalgaard said the
offending code also causes R to come to a stop using SUSE + WINE. Is it
possible to run that lot on top of valgrind? Of course, it will probably
take all day ...
If not, I have a clue which might help. The problem seems to lie in the
"sub" routine. In the original report I used
-- cut here --
gctorture()
u <- intToUtf8(c(rep(1e3,1e2),32,c(rep(1e3,1e2))))
v <- rep(u,1e2)
v <- sub(" ","",v)
v %in% ""
-- cut here --
I've tried reducing this a bit more. Replacing intToUtf8 with a direct
assignment writing out the string with Unicode escapes seems to make no
difference. The %in% can be replaced with "match", leaving the following:
-- cut here --
gctorture()
u <- intToUtf8(c(rep(1e3,1e2),32,c(rep(1e3,1e2))))
v <- rep(u,1e2)
v <- sub(" ","",v)
match(v,"")
-- cut here --
This also crashes R-2.10.0 and R-2.10.1 RC (2009-12-06 r50684).
The sub line is essential, so far as I can see, without it we don't get
the crash. If we add "perl = TRUE" this seems to make no difference (there
is still a crash). If instead we use "fixed = TRUE", the result is strange
and differs for R-2.10.0 and R-2.10.1 RC. This is especially strange,
because in an unbugged R, the result of v returned from sub should be the
same either with fixed = TRUE or perl = TRUE.
R-2.10.0 pauses several seconds, then produces the enigmatic output:
match(v,"")
[1] 00 00 06 9d 78 9c cd 54 5d 4f 83 30 14 2d ec 9b a9 33 99 2f fe 89 65
1a e3
[26] c3 de 8c 26 be 38 7d d5 c7 4a af 0c 57 ca 42 cb 8c bf dc 18 93 61 29
1d 83
[51] 8e 7d c4 18 23 49 a1 f4 de 9e 7b 4e ef 81 47 07 21 64 23 5b 3e ec 1a
42 56
[76] 4b de ea b6 5c b3 e4 e8 c8 d1 f2 80 41 e4 bb 32 7e 5c 58 ae f3 49 f8
66 a6
[101] ce b0 3b c5 1e c8 69 31 b5 15 80 98 84 84 cb e9 22 db 61 27 1b 53 4a
80 0d
[126] 2f 0a e3 99 9c f4 91 ba 4a 41 67 8e 69 0c d7 14 73 ae d1 cc 8c 0e f7
3d 86
[151] 45 1c 81 41 be 19 3e bf 82 2b 4c 40 ee 07 33 0a 0f 8c be a7 6f 3a 62
ad 68
[176] af e8 12 78 c1 31 15 c8 aa 9d 9a a1 5c 89 dd 57 cb eb 27 da 14 38 f2
20 dd
[201] 5c 45 ba c1 70 00 9b 14 35 dc 4c 6e 49 4d 51 e6 8e d3 4d 95 54 aa f1
19 90
[226] 32 21 27 29 73 e8 26 bf 53 d6 32 71 0a 4e da 01 51 49 e3 84 48 77 ce
81 dc
[251] 64 2d 19 ab 0d 7b 33 42 9f 99 42 64 af a7 a8 5e 9d 0f ce 86 83 a1 d9
c1 a5
[276] 7f d0 97 86 69 16 a2 dd 54 90 a6 93 21 1f 25 ba c2 d2 54 68 25 c8 31
49 d6
[301] ae ee 9f 2a ba d4 96 a6 89 03 60 22 83 2b 3b 17 53 3a ce 79 17 3e 96
b5 d3
[326] ea ea b4 3b 9f 8b 9f b9 a5 cd a7 40 41 84 4c a9 ce dd dd 49 fe c6 3e
07 7f
[351] 54 e7 7f d9 f4 50 77 5c 19 a9 e0 b9 5e b2 dd 60 79 6c 13 ad 1e 17 98
11 1c
[376] 91 db e5 5f 7e df 0f e7 43 57 c9 7d 01 6c 3e 1a 5d 0c 2f ab 99 6e a9
a8 88
[401] 57 1c 35 5a 7c 03 73 22 e4 b1
R-2.10.1 RC produces the following equally enigmatic output:
match(v,"")
NULL
Fehler: 'getEncChar' muss f?r CHARSXP aufgerufen werden
So my provisional guess is the bug is somewhere in the part of the
internal code for sub which is invoked whatever the value of fixed or
perl. It is strange though that it makes a difference whether you specify
fixed = TRUE or not.
George Russell
Prof Brian Ripley <ripley at stats.ox.ac.uk> schrieb am 10.12.2009 08:00:36:
It seems (from the debugger output) that this is corruption in the R
memory allocation routines. Such things can usually be tracked down
via valgrind and a valgrind-instrumented build of R, but I cannot
trigger this on any system with valgrind. I've tried 64- and 32-bit
versions, and Latin-1 locales as well as UTF-8.
So I am inclining to think this is Windows-specific. One thing that
is specific to Windows is UCS-2 (16-bit) wide characters, which might
be the issue. But we simply don't have the tools on Windows that we
do on other platforms.
On Wed, 9 Dec 2009, g.russell at eos-solutions.com wrote:
Hello Peter,
I have now installed R-2.10.1 RC (sessionInfo() says "R version 2.10.1
RC
(2009-12-06 r50684)", the rest I believe is as before). The following
code
always brings R --vanilla down (with a crash, not a normal exit):
-- cut here --
gctorture()
u <- intToUtf8(c(rep(1e3,1e2),32,c(rep(1e3,1e2))))
v <- rep(u,1e2)
v <- sub(" ","",v)
v %in% ""
q()
-- cut here --
I've tried this several times now, with different effects. Sometimes R
crashes after 'v %in% ""'. Sometimes it survives that command, but
crashes
during the q(). I have also had the error message "Fehler in match(x,
table, nomatch = 0L) > 0L : Vergleich (6) ist nur f?r atomare und
Listentypen m?glich" from that command (the match seems to be the
problem), when I type q() R still crashes.
Best wishes,
George Russell | KG EOS Holding GmbH & Co
Tel: +49 40 2850 ? 1574 | g.russell at eos-solutions.com
EOS. With head and heart in finance
KG EOS Holding GmbH & Co | Steindamm 71, 20099 Hamburg | AG Hamburg
HRA 95
748
Pers?nlich haftend | EOS Holding GmbH | AG Hamburg HRB 78 748
Gesch?ftsf?hrer | Hans-Werner Scherer, Klaus Engberding, Justus
Hecking-Veltman, Paul Leary sen., Christos Savvides, Dr. Andreas
Witzig
Vorsitzender des Beirates | J?rgen Schulte-Laggenbeck
Save a tree. Don?t print this email unless it?s really necessary.
Diese E-Mail enth?lt vertrauliche und/oder rechtlich gesch?tzte
Informationen. Wenn Sie nicht der richtige Adressat sind oder diese
E-Mail
irrt?mlich erhalten haben, informieren Sie bitte sofort den Absender
und
vernichten Sie diese Mail. Das unerlaubte Kopieren sowie die unbefugte
Weitergabe dieser Mail ist nicht gestattet.
This email may contain confidential and/or privileged information.
If you are not the intended recipient or have received this email in
error, please notify the sender immediately and destroy this email.
Any unauthorized copying, disclosure or distribution of the material
in
this email is strictly forbidden.
Peter Dalgaard <P.Dalgaard at biostat.ku.dk> schrieb am 08.12.2009
11:24:50:
g.russell at eos-solutions.com wrote:
Full_Name: George Russell
Version: 2.10.0
OS: Windows XP Version 2002 SP 2
Submission from: (NULL) (217.111.3.131)
The following typed into R --vanilla induces a crash:
-- cut here --
gctorture()
u <- intToUtf8(c(rep(1e3,1e2),32,c(rep(1e3,1e2))))
v <- rep(u,1e2)
v <- sub(" ","",v)
v %in% ""
-- cut here --
sessionInfo() says:
-- cut here --
R version 2.10.0 (2009-10-26)
i386-pc-mingw32
locale:
[1] LC_COLLATE=German_Germany.1252 LC_CTYPE=German_Germany.1252
[3] LC_MONETARY=German_Germany.1252 LC_NUMERIC=C
[5] LC_TIME=German_Germany.1252
attached base packages:
[1] stats graphics grDevices datasets utils methods base
-- cut here --
I apologise for not testing this with R-2.10.1 but as far as I can
see there are
only source releases available so far, which I am not able to
compile.
2.10.1 RC is available now. Please check. It does seem to be
reproducible in the Windows version, or at least it takes a very long
time, but that means running under Wine on SUSE for me. I don't see
the
effect with the Linux build.
--
O__ ---- Peter Dalgaard ?ster Farimagsgade 5, Entr.B
c/ /'_ --- Dept. of Biostatistics PO Box 2099, 1014 Cph. K
(*) \(*) -- University of Copenhagen Denmark Ph: (+45)
35327918
~~~~~~~~~~ - (p.dalgaard at biostat.ku.dk) FAX: (+45)
--
Brian D. Ripley, ripley at 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