Hi again!
While investigating the bug report [*] I found out that on unix, system2
does not quote its `stdin` argument while preparing the command line to launch.
It does shQuote the `stdout` and `stderr` arguments, and also the `f <-
tmpfile()` variable (which is used if `input` argument is provided),
which seems to set a precedent. On Windows, stdin, stdout, and stderr are handled independently of the shell, so it also just works without the use of shQuote by the caller.
Have people been relying on system2 not quoting the `stdin` argument, but
quoting `stdout` and `stderr`? For what it's worth, neither R_runR in
src/library/tools/R/check.R, nor .system_with_capture in
src/library/tools/R/utils.R (the only callers of system2(..., stdin = ...)),
nor their callers seem to be shQuote'ing the `stdin` argument.
Nor the rare system2(..., stdin = ...) callers (or their callers) on CRAN
seem to be quoting the `stdin` argument (I did find one exception [**]), it usually being "" or tmpfile() passed across a few function calls.
Given the considerations above, would the following patch be a good idea?
Index: src/library/base/R/unix/system.unix.R
===================================================================
--- src/library/base/R/unix/system.unix.R (revision 77566)
+++ src/library/base/R/unix/system.unix.R (working copy)
@@ -102,7 +102,7 @@
writeLines(input, f)
## here 'command' is a single command, unlike system()
command <- paste(command, "<", shQuote(f))
- } else if (nzchar(stdin)) command <- paste(command, "<", stdin)
+ } else if (nzchar(stdin)) command <- paste(command, "<", shQuote(stdin))
if(!wait && !intern) command <- paste(command, "&")
.Internal(system(command, intern, timeout))
}