pure/impure?!?!

[David Kastrup]

In the CG I read

    For certain grobs, the @code{Y-extent} is based on the @code{stencil}
    property, overriding the stencil property of one of these will
    require an additional @code{Y-extent} override with an unpure-pure
    container.  When a function overrides a @code{Y-offset} and/or
    @code{Y-extent} it is assumed that this will trigger line breaking
    calculations too early during compilation.  So the function is not
    evaluated at all (usually returning a value of @samp{0} or
    @samp{'(0 . 0)}) which can result in collisions.  A @q{pure} function
    will not affect properties, objects or grob suicides and therefore will
    always have its Y-axis-related evaluated correctly.

    Currently, there are about thirty functions that are already considered
    @q{pure} and Unpure-pure containers are a way to set functions not on
    this list as @q{pure}.  The @q{pure} function is evaluated @emph{before}
    any line-breaking and so the horizontal spacing can be adjusted
    @q{in time}.  The @q{unpure} function is then evaluated @emph{after}
    line breaking.

This is all very nice.  Except that it is the function calls labelled as
*pure* that actually receive start/end values indicating particular
breakpoints of the system that the respective grob is in.  What's the
deal with that?

-- 
David Kastrup