base (version 3.5.2)

tracemem: Trace Copying of Objects


This function marks an object so that a message is printed whenever the internal code copies the object. It is a major cause of hard-to-predict memory use in R.


retracemem(x, previous = NULL)



An R object, not a function or environment or NULL.


A value as returned by tracemem or retracemem.


A character string for identifying the object in the trace output (an address in hex enclosed in angle brackets), or NULL (invisibly).


This functionality is optional, determined at compilation, because it makes R run a little more slowly even when no objects are being traced. tracemem and untracemem give errors when R is not compiled with memory profiling; retracemem does not (so it can be left in code during development).

It is enabled in the CRAN macOS and Windows builds of R.

When an object is traced any copying of the object by the C function duplicate produces a message to standard output, as does type coercion and copying when passing arguments to .C or .Fortran.

The message consists of the string tracemem, the identifying strings for the object being copied and the new object being created, and a stack trace showing where the duplication occurred. retracemem() is used to indicate that a variable should be considered a copy of a previous variable (e.g., after subscripting).

The messages can be turned off with tracingState.

It is not possible to trace functions, as this would conflict with trace and it is not useful to trace NULL, environments, promises, weak references, or external pointer objects, as these are not duplicated.

These functions are primitive.

See Also

capabilities("profmem") to see if this was enabled for this build of R.

trace, Rprofmem


Run this code
a <- 1:10
## b and a share memory
b <- a
b[1] <- 1

## copying in lm: less than R <= 2.15.0
d <- stats::rnorm(10)
lm(d ~ a+log(b))

## f is not a copy and is not traced
f <- d[-1]
## indicate that f should be traced as a copy of d
retracemem(f, retracemem(d))
# }

Run the code above in your browser using DataLab