Version 7 of empty string

Updated 2014-05-29 21:08:21 by AMG

PYK: The empty string is a string containing no characters. Because of Tcl's copy-on-write implementation of values, there is precisely one empty string. In a Tcl script, there are exactly three notations for the empty string: {}, "", and [some_command_that_returns_the_empty_string_including_possibly_no_command_at_all].

AMG: Actually, you can obtain empty string by typing [].

PYK :P

AMG: It is possible for multiple Tcl_Objs to be empty string, though this fact is completely invisible at the script level except when probed using [tcl::unsupported::representation]. (By the way, making decisions based on the result of that command is forbidden except for development, debugging, and profiling purposes. tcom's unpredictable type mapping is a great example of what can happen should you ignore this requirement.)

% info patchlevel
8.6.1
% tcl::unsupported::representation {}
value is a list with a refcount of 28, object pointer at 0000000002725AF0, internal representation 0000000002774150:0000000000000000, string representation ""
% tcl::unsupported::representation []
value is a list with a refcount of 29, object pointer at 0000000002725AF0, internal representation 0000000002774150:0000000000000000, string representation ""
% tcl::unsupported::representation [list]
value is a list with a refcount of 30, object pointer at 0000000002725AF0, internal representation 0000000002774150:0000000000000000, string representation ""
% tcl::unsupported::representation ""
value is a list with a refcount of 31, object pointer at 0000000002725AF0, internal representation 0000000002774150:0000000000000000, string representation ""
% tcl::unsupported::representation [set var ""]
value is a list with a refcount of 33, object pointer at 0000000002725AF0, internal representation 0000000002774150:0000000000000000, string representation ""
% tcl::unsupported::representation [append var]
value is a list with a refcount of 32, object pointer at 0000000002725AF0, internal representation 0000000002774150:0000000000000000, string representation ""
% tcl::unsupported::representation [append var ""]
value is a list with a refcount of 2, object pointer at 0000000002A6A8D0, internal representation 0000000002774150:0000000000000000, string representation ""

In this case, [append]ing empty string to a variable whose value is (shared) empty string caused a copy to be made even though the resultant value was identical to the original. In other words, [append] is not optimized to detect when all its value arguments are empty string, since that is a very rare use.

Have a look at this interesting case demonstrating shimmering and value sharing:

% tcl::unsupported::representation [set var ""]
value is a list with a refcount of 27, object pointer at 0000000002615AF0, internal representation 0000000002663BD0:0000000000000000, string representation ""
% regexp $var x
1
% tcl::unsupported::representation ""
value is a regexp with a refcount of 30, object pointer at 0000000002615AF0, internal representation 00000000029140C0:0000000000000000, string representation ""

Here, the intrep of empty string changed from list to regexp. But so what? It can freely change back. There's no semantic consequence. This is just an implementation curiosity.


See also: is_empty