Macro sets from the earliest days of TeX programming may be observed to test whether commands exist by using
(which of course actually tests that the command doesn’t exist). LaTeX programmers can make use of the internal command\ifx\command\undefined‹stuff› …
which executes action1 if the command is undefined, and action2 if it is defined (cmd name is the command name only, omitting the ‘\@ifundefined{cmd name}{action1}{action2}
\’ character).
The \@ifundefined command is based on the sequence
which relies on the way\expandafter \ifx \csname cmd name\endcsname \relax
\csname works: if the command doesn’t
exist, it simply creates it as an alias for \relax.
So: what is wrong with these techniques?
Using \undefined blithely assumes that the command is indeed not
defined. This isn’t entirely safe; one could make the name more
improbable, but that may simply make it more difficult to spot a
problem when things go wrong. LaTeX programmers who use the
technique will typically employ \@undefined, adding a single
level of obscurity.
The \@ifundefined mechanism has the unfortunate property of
polluting the name space: each test that turns out undefined adds a
name to the set TeX is holding, and often all those “\relax”
names serve no purpose whatever. Even so (sadly) there are places in
the code of LaTeX where the existence of the \relax is relied
upon, after the test, so we can’t get away from \@ifundefined
altogether.
David Kastrup offers the (rather tricky)
{\expandafter}\expandafter\ifx \csname cmd name\endcsname\relax ...
which “creates” the \relax-command inside the group of the first
\expandafter, therefore forgets it again once the test is done.
The test is about as good as you can do with macros.
The e-TeX system system comes to our help here: it defines two new primitives:
\ifdefined, which tests whether a thing is defined (the
negative of comparing with \undefined, as it were), and
\ifcsname cmd name\endcsname, which does the
negative of \@ifundefined without the \relax-command
side-effect.
\ifdefined\foo
\message{\string\foo\space is defined}%
\else
\message{no command \string\foo}%
\fi
%
\ifcsname foo\endcsname
\message{\string\foo\space is defined}%
\else
\message{no command \string\foo}%
\fi
However, after using the LaTeX
\@ifundefined{foo}…, the conditionals will detect the
command as “existing” (since it has been \let to \relax);
so it is important not to mix mechanisms for detecting the state of a
command.
Since most distributions nowadays use e-TeX as their base executable for most packages, these two primitives may be expected appear widely in new macro packages.
This question on the Web: http://www.tex.ac.uk/cgi-bin/texfaq2html?label=isdef