Heiko’s answer is a little terse, so I hope that I will be allowed to add a few words of explanation, even if I do not propose an alternative solution.
I would say that the answer to your question is “it depends on whether the outer switch is true or false”. I’ll analyze your second example, that deals with \ifA and \ifB, in order to explain why Heiko’s answer works.
If TeX finds that the outer \ifA is false, it will begin to skip tokens until it finds the corresponding \else or \fi; while doing so, if it runs into a control sequence that has already been defined as a conditional, it keeps track that it starts a new, nested conditional. But when TeX has begun to skip tokens, that is, immediately after the \ifA, the control sequence \ifB was not yet known to represent a conditional; thus, TeX simply skips over it like over any other token, and when it comes to the ensuing \else, it thinks that this is the \else it was looking for, that is, the one which goes with \ifA, and so it resumes executing tokens. The next token is \fi, which is interpreted as the end of the \ifA conditional. As you can now see, the following \else is obviously misplaced in TeX’s eyes.
On the other hand, everything goes fine if the outer \ifA switch is true, because in this case TeX executes the definition of \ifB before seeing it. You can try this yourself:
\documentclass{article}
\newif\ifA
\Atrue
\ifA
\newif\ifB
\ifB
\else
\fi
\else
\fi
\begin{document}
AB
\end{document}
The reason why Heiko’s answer works is, of course, that TeX always knows that \ifx is a conditional. So, even if it is skipping over the “true branch” of \ifA, it can tell that the first \else, and the ensuing \fi, should be paired with \ifx and not with \ifA.
Addition
A comment suggests that I should further clarify the matter.
Yes, the usual paradigms you may be accustomed to when thinking of programming languages aren’t appropriate for TeX: TeX is different, and it probably shouldn’t be even considered a programming language, but rather a purely sequential processor equipped with a very powerful macro preprocessor; with a crucial feature, however: that the meaning of macros can be defined, and changed, at run-time.
Coming to our specific case, there is no built-in “syntactic” rule that dictates that control sequences starting with \if... are conditionals: this is merely a convention that should be adhered to, but nothing more. (Note, however, that plain TeX’s \newif enforces it; LaTeX2e’s doesn’t, but assumes that the “core” of the name of the conditional is preceded by a two-letter prefix.) As far as the “machine level” of TeX is concerned, there are only a finite number of conditionals: for Knuth’s original TeX3, they are the 14 control sequences listed on pages 209–210 of The TeXbook, namely
\ifnum,
\ifdim,
\ifodd,
\ifvmode,
\ifhmode,
\ifmmode,
\ifinner,
\if,
\ifcat,
\ifx,
\ifvoid,
\ifeof,
\iftrue, and
\iffalse
(there is also the \ifcase primitive, but it is quite different from the others). Other typesetting engines (e-TeX, pdfTeX, and so on) add many other primitive conditionals to this list, each of them having its peculiar list of additions, but it remains true that only the control sequences explicitly belonging to these finite lists of primitives can act as a conditional, and in particular are recognized as such when skipping over tokens belonging to a branch of an (outer) conditional that hasn’t been taken.
How can we define new, “custom” conditionals, then? Well, as already said above, the macro processor included in TeX has the peculiar characteristic that the meaning of macros can be changed at run-time, and this is how the ability to define new conditionals is attained. Let’s see in detail how this works by means of an example.
When you say
\newif\ifFOO
the following things happen:
First, the equivalent of
\let\ifFOO=\iffalse
is executed. So, the control sequence \ifFOO is made a perfect
synonym of \iffalse; in particular, from now on TeX will recognize
\ifFOO as a conditional (unless, or until, it is redefined,
of course), even when skipping over tokens at high speed.
Two new control sequences \FOOtrue and \FOOfalse are defined,
with a meaning equivalent to that given them by the following code:
\def\FOOfalse{\let\ifFOO=\iffalse}
\def\FOOtrue {\let\ifFOO=\iftrue}
So, you can now say, for example
\FOOtrue
in some place of your code, and that will make subsequent occurrences of \ifFOO behave exactly as \iftrue had been written in their place; and similarly for \FOOfalse.
I hope that this suffices to clarify how \if... conditionals work.
