Error when calling (R 4.0.x on Windows) from Python
R/Rterm now has shell-independent ways to specify its standard input file, --file=file.R, and environment variables, VAR=VALUE. Should it also have shell-independent arguments to specify files to contain stdout and stderr or both? Then its help message could omit the '> output' (it can already leave out the '< infile'). The help message could then say that if those arguments are omitted then Rterm uses stdin, stdout, and stderr. Thus the user could come up with the shell syntax if the user needed to use pipes or output stream merging, etc. -Bill
On Tue, Feb 2, 2021 at 5:49 AM Tomas Kalibera <tomas.kalibera at gmail.com> wrote:
A reproducible example:
system("\"C:\\Program Files\\R\\R-4.0.3\\bin\\R.exe\" -e commandArgs()
>out") # does not create "out"
system("\"C:\\Program Files\\R\\R-3.6.3\\bin\\R.exe\" -e commandArgs()
>out") # creates "out"
Note that this does not create "out", either:
system("\"C:\\Program Files\\R\\R-3.6.3\\bin\\x64\\Rterm.exe\" -e
commandArgs() >out")
I think redirections should be left to the shell, so e.g. those calls
from Python should use "os.system" if they needed redirection, as Duncan
suggested.
I see that both R.exe and Rterm.exe give "Usage: Rterm [options] [<
infile] [> outfile] [EnvVars]", which may be confusing to programmers
invoking R without a shell, but then talking there about invocation via
shell could confuse typical users, instead.
My best guess is that the redirection in R.exe in 3.6 worked rather by
accident, as a consequence of that R.exe internally invoked (and still
invokes) Rterm.exe via the shell, but R.exe did not protect all
arguments from expansions from that internal shell invocation. If
redirection was meant to work without external shell, it would have
instead been implemented explicitly also in Rterm.exe.
My change 77926 made sure that all arguments to that internal shell
invocation were protected, following bug reports such as PR#17709, where
users had "&" in their file names. Therefore, this works, executing file
code&.r
system("\"C:\\Program Files\\R\\R-4.0.3\\bin\\R.exe\" -f code&.r")
but not with 3.6.3, and I would not be surprised if even more special
characters became popular, soon.
After all, what if someone wanted to pass an argument to R including
">", that should work, too:
---- t.r ----
cat("Header: ", commandArgs(TRUE)[1], "\n")
-------------
system("\"C:\\Program Files\\R\\R-4.0.3\\bin\\R.exe\" -f t.r --args
<html>") # prints Header: <html>
system("\"C:\\Program Files\\R\\R-3.6.3\\bin\\R.exe\" -f t.r --args
<html>") # fails with The syntax of the command is incorrect.
So in summary, I don't agree this is a bug.
Best
Tomas
On 1/29/21 11:19 AM, Duncan Murdoch wrote:
On 29/01/2021 3:57 a.m., Marcel Baumgartner wrote:
Dear Bill, Duncan and Martin, thanks for your investigation. Can you clarify on next steps? Is this now an official bug, or have you found a workaround? For your information: the issue showed up the first time when I called R 4.0.2 from within a software called "IDEA" (from Caseware Analytics), using their scripting language (similar to Visual Basic). With my colleague we then simply reproduce the error calling R from Python, so that we could share it more easily. When we run this command directly on the CMD in Windows, all works fine. The issue only happens when R is called within another software.
I would say this is a bug, with two workarounds. These are only
needed (and will only work) on Windows.
1. If you want to include redirection in the command line, then call
cmd.exe yourself, and have it call Rterm.exe. I think your original
command
R.exe -f code.R --args "~/file.txt" 1> "~/log.txt" 2>&1"
could be written as
cmd.exe 1>log.txt 2>&1 /C R.exe -f code.R --args "~/file.txt"
(but I haven't tested it). You could also use the slightly more
efficient
cmd.exe 1>log.txt 2>&1 /C Rterm.exe -f code.R --args "~/file.txt"
if Rterm.exe is on your path; by default I don't think it will be.
2. Change the Python code to a different function call than
subprocess.call. It should be one that's equivalent to the C system()
function; I believe os.system() is what you want, so you'd use
os.system("R.exe -f code.R --args "~/file.txt" 1> "~/log.txt" 2>&1")
and the redirection would be handled by the implicit shell that is
called by os.system. Again, replacing R.exe by Rterm.exe would be a
tiny bit more efficient, but it might not be on the path.
Duncan Murdoch
Best regards
Marcel
Le 2021-01-27T23:14:36.000+01:00, Bill Dunlap
<williamwdunlap at gmail.com> a ?crit :
I tried the following change, that adds quotes if the argument does
not include ">".
Index: front-ends/rcmdfn.c
===================================================================
--- front-ends/rcmdfn.c (revision 79883)
+++ front-ends/rcmdfn.c (working copy)
@@ -173,9 +173,13 @@
fprintf(stderr, "command line too long\n");
return(27);
}
- strcat(cmd, "\"");
+ if (!strchr(argv[i], '>')) {
+ strcat(cmd, "\"");
+ }
strcat(cmd, argv[i]);
- strcat(cmd, "\"");
+ if (!strchr(argv[i], '>')) {
+ strcat(cmd, "\"");
+ }
}
/* the outermost double quotes are needed for cmd.exe */
strcat(cmd, "\"");
It lets the python example work. I am not sure that quoting all the
arguments buys you much, as shQuote() is still needed for arguments
that include spaces. E.g., with 3.6.3, 4.0.3, and my development
build with the above patch we get
stopifnot(dir.create(dirname <- file.path(tempfile(), "A
SPACE"), recursive=TRUE))
logname <- file.path(dirname, "log.txt")
unlink(logname)
system(paste( "C:\\R\\R-3.6.3\\bin\\R.exe --quiet --vanilla -e
\"commandArgs()\" 1>", logname))
ARGUMENT 'SPACE/log.txt' __ignored__
[1] 0
tryCatch(readLines(logname),
error=function(e)conditionMessage(e))
[1] "cannot open the connection"
Warning message:
In file(con, "r") :
cannot open file
'C:\Users\willi\AppData\Local\Temp\RtmpM5tsC7\file1a1068734a49/A
SPACE/log.txt': No such file or directory
system(paste( "C:\\R\\R-4.0.3\\bin\\R.exe --quiet --vanilla -e
\"commandArgs()\" 1>", logname))
commandArgs()
[1] "C:\\R\\R-40~1.3/bin/x64/Rterm.exe"
[2] "--quiet"
[3] "--vanilla"
[4] "-e"
[5] "commandArgs()"
[6] "1>"
[7]
"C:\\Users\\willi\\AppData\\Local\\Temp\\RtmpM5tsC7\\file1a1068734a49/A"
[8] "SPACE/log.txt"
[1] 0
tryCatch(readLines(logname),
error=function(e)conditionMessage(e))
[1] "cannot open the connection"
Warning message:
In file(con, "r") :
cannot open file
'C:\Users\willi\AppData\Local\Temp\RtmpM5tsC7\file1a1068734a49/A
SPACE/log.txt': No such file or directory
unlink(logname)
system(paste(
"C:\\msys64\\home\\willi\\ucrt3\\r\\trunk\\bin\\R.exe --quiet
--vanilla -e \"commandArgs()\" 1>", logname))
[1] 0
tryCatch(readLines(logname),
error=function(e)conditionMessage(e))
[1] "cannot open the connection"
Warning message:
In file(con, "r") :
cannot open file
'C:\Users\willi\AppData\Local\Temp\RtmpM5tsC7\file1a1068734a49/A
SPACE/log.txt': No such file or directory
tryCatch(readLines(sub(" .*$", "", logname)),
error=function(e)conditionMessage(e))
[1] "> commandArgs()"
"[1]
\"C:\\\\msys64\\\\home\\\\willi\\\\ucrt3\\\\r\\\\trunk/bin/x64/Rterm.exe\""
[3] "[2] \"--quiet\"
" "[3] \"--vanilla\"
"
[5] "[4] \"-e\"
" "[5] \"commandArgs()\"
"
[7] "[6] \"SPACE/log.txt\"
" "> "
[9] "> "
unlink(logname)
system(paste(
"C:\\msys64\\home\\willi\\ucrt3\\r\\trunk\\bin\\R.exe --quiet
--vanilla -e \"commandArgs()\" 1>", shQuote(logname)))
[1] 0
tryCatch(readLines(logname),
error=function(e)conditionMessage(e))
[1] "> commandArgs()"
"[1]
\"C:\\\\msys64\\\\home\\\\willi\\\\ucrt3\\\\r\\\\trunk/bin/x64/Rterm.exe\""
[3] "[2] \"--quiet\"
" "[3] \"--vanilla\"
"
[5] "[4] \"-e\"
" "[5] \"commandArgs()\"
"
[7] "> "
"> "
-Bill
On Wed, Jan 27, 2021 at 1:25 PM Duncan Murdoch
<murdoch.duncan at gmail.com <mailto:murdoch.duncan at gmail.com>> wrote:
On 27/01/2021 3:40 p.m., Bill Dunlap wrote:
I believe the problem is from svn 77925 in
gnuwin/front-ends/rcmdfn.c,
which was committed a few days after 3.6.3 was released.
Rterm used
to put double quotes around a command line argument only
if it
contained a space, now it double quotes all arguments. It
sees shell
constructs like "1>" and the following file name as
arguments and
double quoting them hides them from the shell, leading to
this
problem. I think we may have to rely on the user supplying
quotes as
needed instead of blindly adding them.
Okay, now I see what you mean.
If you invoke R using R.exe, it asks cmd.exe to run Rterm.exe,
so it is
possible that redirection would be handled.
If you invoke R directly using Rterm.exe, then my description
down below
would be correct.
Duncan Murdoch
-Bill
On Wed, Jan 27, 2021 at 12:28 PM Duncan Murdoch
<murdoch.duncan at gmail.com <mailto:murdoch.duncan at gmail.com>>
wrote:
On 27/01/2021 3:17 p.m., Duncan Murdoch wrote:
On 27/01/2021 3:38 a.m., Martin Maechler wrote:
Martin Maechler
on Tue, 26 Jan 2021 12:37:58
+0100 writes:
Marcel Baumgartner
on Tue, 26 Jan 2021 08:55:48
+0100 writes:
Dear all, my colleague posted our issue on
stackoverflow:
Calling R script from Python does not save log
file in
version 4 - Stack Overflow
[stackoverflow.com/questions...
<https://stackoverflow.com/questions/65887485/calling-r-script-from-python-does-not-save-log-file-in-version-4>]
It is about this kind of call to R:
R.exe -f code.R --args "~/file.txt" 1>
"~/log.txt" 2>&1".
The issue is that the log.txt file is not
created when
running R 4.x.x. The same code works perfectly
fine with
R 3.6.x.
Any idea what's going wrong as of version 4?
Regards
Marcel
Dear Marcel, I think the solution is
embarrassingly
simple:
From the SO post, where she showed a bit more
detail than you
show here, it's clear you have confused
'R.exe' and
'Rscript.exe' and what you say above is not
true:
'R.exe' was used for R 3.6.0 but for R 4.0.3,
you/she used
'Rscript.exe' instead.
... as you've noticed now, they do behave
differently,
indeed!
Well, this was not the solution to their --
Windows-only -- problem.
The problem *is* indeed visible if they only use
R.exe (also
for R 4.0.3).
I've commented more on the SO issue (see above),
notably asking for a *minimal* repr.ex.
(reproducible example),
and one *not* using "<YOUR PATH>" and setwd() ..
Isn't this purely a Python or user problem? R
shouldn't process
redirection directives like
1> "~/log.txt" 2>&1
because it's the shell's job to process those. If
Python is acting as
the shell, it needs to handle those things. If R was
handling the
command via
Oops, sent before finishing:
If R was handling the command via system() or system2(),
it would handle
redirection itself. If it was using the Windows-only
shell(), it would
call cmd.exe (by default) to handle redirection. (This
is a difference
between R on Windows and R in Unix: in Unix a shell is
always used.)
Duncan Murdoch
______________________________________________
R-help at r-project.org <mailto:R-help at r-project.org>
mailing list -- To UNSUBSCRIBE and more, see
stat.ethz.ch/mailman/listin...
<https://stat.ethz.ch/mailman/listinfo/r-help>
PLEASE do read the posting guide
www.R-project.org/posting-g...
<http://www.R-project.org/posting-guide.html>
and provide commented, minimal, self-contained,
reproducible code.
______________________________________________ R-help at r-project.org mailing list -- To UNSUBSCRIBE and more, see https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code.