[Rd] Rscript -e EXPR fails to launch if stdin is closed

[peter dalgaard]

There's still 2 weeks till code freeze for 4.2.2, and porting the fix would be trivial. As long as there is no risk that someone will get the bright idea of changing a critical package to depend on R >= 4.2.2...

- pd

> On 10 Oct 2022, at 18:34 , Henrik Bengtsson <henrik.bengtsson using gmail.com> wrote:
> 
> Thank you Peter for the quick fix.  Will this make it into R-patched
> to become R 4.2.2 soon?
> 
> I can confirm that the fix resolved also the original problem report
> involving launching a parallel PSOCK cluster from within a 'processx'
> background process
> (https://stackoverflow.com/questions/73962109/why-are-the-workers-failing-to-connect-when-calling-makepsockcluster-from-an-e/73991833#73991833
> and https://github.com/r-lib/callr/issues/236)
> 
> 
> /Henrik
> 
> On Mon, Oct 10, 2022 at 5:54 AM peter dalgaard <pdalgd using gmail.com> wrote:
>> 
>> It seems to work simply to do  "if (ifd >= 0)..." (the ifp test is fine since ifp is FILE* and initialized to NULL). Will commit (to r-devel for now).
>> 
>> -pd
>> 
>>> On 10 Oct 2022, at 11:07 , peter dalgaard <pdalgd using gmail.com> wrote:
>>> 
>>> He!
>>> 
>>> Yes, that looks like a blunder.
>>> 
>>> mkstemp() returns -1 on failure, not 0, so the test on ifd (and I suppose also the one on ifp) is wrong. And of course, once you close file descriptor 0, mkstemp() chooses the 1st available fd, i.e. 0, for its return value.
>>> 
>>> -pd
>>> 
>>>> On 9 Oct 2022, at 20:25 , Henrik Bengtsson <henrik.bengtsson using gmail.com> wrote:
>>>> 
>>>> Rscript fails to launch if the standard input (stdin) is closed, e.g.
>>>> 
>>>> $ Rscript --vanilla -e 42 0<&-
>>>> Fatal error: creating temporary file for '-e' failed
>>>> 
>>>> This appear to only happen with `-e EXPR`, e.g. it works when doing:
>>>> 
>>>> $ echo "42" > script.R
>>>> $ Rscript --vanilla script.R 0<&-
>>>> [1] 42
>>>> 
>>>> and:
>>>> 
>>>> $ R --vanilla 0<&-
>>>> R version 4.2.1 (2022-06-23) -- "Funny-Looking Kid"
>>>> Copyright (C) 2022 The R Foundation for Statistical Computing
>>>> Platform: x86_64-pc-linux-gnu (64-bit)
>>>> ...
>>>>> 
>>>> 
>>>> 
>>>> TROUBLESHOOTING:
>>>> 
>>>> $ strace Rscript --vanilla -e 42 0<&-
>>>> execve("/home/hb/shared/software/CBI/R-4.2.1-gcc9/bin/Rscript",
>>>> ["Rscript", "--vanilla", "-e", "42"], 0x7fff9f476418 /* 147 vars */) =
>>>> 0
>>>> brk(NULL)                               = 0x5625ca9e6000
>>>> arch_prctl(0x3001 /* ARCH_??? */, 0x7fff23b4d260) = -1 EINVAL (Invalid argument)
>>>> ...
>>>> fstat(1, {st_mode=S_IFCHR|0620, st_rdev=makedev(0x88, 0), ...}) = 0
>>>> write(1, "Fatal error: creating temporary "..., 53Fatal error:
>>>> creating temporary file for '-e' failed
>>>> ) = 53
>>>> exit_group(2)                           = ?
>>>> +++ exited with 2 +++
>>>> 
>>>> which points to src/unix/system.c:
>>>> 
>>>> ifd = mkstemp(ifile);
>>>> if (ifd > 0)
>>>>  ifp = fdopen(ifd, "w+");
>>>> if(!ifp) R_Suicide(_("creating temporary file for '-e' failed"));
>>>> 
>>>> 
>>>> One rationale for having closed standard files (including stdin) is to
>>>> avoid leaking file descriptors, cf.
>>>> https://wiki.sei.cmu.edu/confluence/display/c/FIO22-C.+Close+files+before+spawning+processes
>>>> and https://danwalsh.livejournal.com/53603.html.  The background for
>>>> reporting on this was that `system()` fails to work in processx
>>>> spawned processes, which closes the standard files by default in
>>>> processx (<= 3.7.0).
>>>> 
>>>> Best,
>>>> 
>>>> Henrik
>>>> 
>>>> ______________________________________________
>>>> R-devel using r-project.org mailing list
>>>> https://stat.ethz.ch/mailman/listinfo/r-devel
>>> 
>>> --
>>> Peter Dalgaard, Professor,
>>> Center for Statistics, Copenhagen Business School
>>> Solbjerg Plads 3, 2000 Frederiksberg, Denmark
>>> Phone: (+45)38153501
>>> Office: A 4.23
>>> Email: pd.mes using cbs.dk  Priv: PDalgd using gmail.com
>>> 
>> 
>> --
>> Peter Dalgaard, Professor,
>> Center for Statistics, Copenhagen Business School
>> Solbjerg Plads 3, 2000 Frederiksberg, Denmark
>> Phone: (+45)38153501
>> Office: A 4.23
>> Email: pd.mes using cbs.dk  Priv: PDalgd using gmail.com
>> 

-- 
Peter Dalgaard, Professor,
Center for Statistics, Copenhagen Business School
Solbjerg Plads 3, 2000 Frederiksberg, Denmark
Phone: (+45)38153501
Office: A 4.23
Email: pd.mes using cbs.dk  Priv: PDalgd using gmail.com