Experimenting a bit further, I've "found" the following
1) the problem seems only gsub(), not sub()
> sub(" ", "", "b c + d | a * b", perl=TRUE)
[1] "bc + d | a * b"
> gsub(" ", "", "b c + d | a * b", perl=TRUE)
NULL
2) only if perl = TRUE, not otherwise
3) Also modifying the replacement string only slightly leads to a
different (correct instead of bug) result:
> gsub(" ", "", "bc + d | a * b", perl=TRUE)
[1] "bc+d|a*b"
> gsub(" ", "", "b c + d | a *", perl=TRUE)
[1] "bc+d|a*"
> gsub(" ", "", "b c + d | b", perl=TRUE)
[1] "bc+d|b"
Whereas those give the bug as well :
> gsub(" ", "", "b : : d : a : b", perl=TRUE)
NULL
> gsub(" ", "", "b : : : : a : b", perl=TRUE)
NULL
> gsub(" ", "", "b : : : : : : b", perl=TRUE)
NULL
> gsub(" ", "", "a : : : : : : a", perl=TRUE)
NULL
but not this
> gsub(" ", "", "b : : : : : : a", perl=TRUE)
[1] "b::::::a"
But it's even worse :
> gsub(" ", "", "a: 12345 :a", perl=TRUE)
-> segmentation fault
and from "R -d gdb" :
> gsub(" ", "", "a: 12345 :a", perl=TRUE)
Program received signal SIGSEGV, Segmentation fault.
hashIndex (symbol=0x8fd0060, table=0x823f2b8) at
../../../R/src/main/envir.c:599
599 if( !HASHASH(c) ) {
(gdb) bt
#0 hashIndex (symbol=0x8fd0060, table=0x823f2b8)
at ../../../R/src/main/envir.c:599
#1 0x080b145d in R_GetGlobalCache (symbol=0x1)
at ../../../R/src/main/envir.c:656
#2 0x080b1ac1 in findGlobalVar (symbol=0x8fd0060)
at ../../../R/src/main/envir.c:923
#3 0x080b8873 in Rf_eval (e=0x823f2b8, rho=0x823f014)
at ../../../R/src/main/eval.c:329
#4 0x080b88c9 in Rf_eval (e=0x8fd1e60, rho=0x821e540)
at ../../../R/src/main/eval.c:354
#5 0x080e1e0e in GetObject (cptr=0xbfffd400)
at ../../../R/src/main/objects.c:88
#6 0x080e2a3b in do_usemethod (call=0x84eb084, op=0x823be78, args=0x84eb0a0,
env=0x8fd1df0) at ../../../R/src/main/objects.c:381
#7 0x080b8c33 in Rf_eval (e=0x84eb084, rho=0x8fd1df0)
at ../../../R/src/main/eval.c:375
#8 0x080b8f48 in Rf_applyClosure (call=0x8fd1eb4, op=0x84eafa4,
arglist=0x8fd1e44, rho=0x823f014, suppliedenv=0x821e540)
at ../../../R/src/main/eval.c:559
#9 0x080b89bf in Rf_eval (e=0x8fd1eb4, rho=0x823f014)
at ../../../R/src/main/eval.c:410
#10 0x0811b0f3 in Rf_PrintValueEnv (s=0x8fd1eb4, env=0x823f014)
at ../../../R/src/main/print.c:775
#11 0x080d55f1 in Rf_ReplIteration (rho=0x823f014, savestack=150798432,
browselevel=1, state=0x823f014) at ../../../R/src/main/main.c:254
#12 0x080d578d in R_ReplConsole (rho=0x823f014, savestack=0, browselevel=0)
at ../../../R/src/main/main.c:298
#13 0x080d60d0 in run_Rmainloop ()
at ../../../R/src/main/main.c:656
#14 0x08147e90 in main (ac=150798432, av=0x8fd0060)
at ../../../R/src/unix/system.c:99
#15 0x42017589 in __libc_start_main () from /lib/i686/libc.so.6
(which is not directly helpful since src/main/pcre.c (where
the C source for gsub(*, perl=TRUE) resides) isn't mentioned above.
Must be a memory allocation / mismatch / end_of_string problem
somewhere.
Martin Maechler, ETH Zurich