How to circumvent a late \pagetotal reset to 0pt?

Question

Below example demonstrates at the beginning of each paragraph how \pagetotal is not always reset to 0pt at the beginning of the page. This appears to interfere with my attempt to automatically clear a floatingfigure. What could be a remedy for this?

\documentclass{article}
\usepackage{mwe}

\makeatletter
\def\maxwidth{\ifdim\Gin@nat@width>\linewidth\linewidth
\else\Gin@nat@width\fi}
\makeatother
\let\Oldincludegraphics\includegraphics
\renewcommand{\includegraphics}[1]{\Oldincludegraphics[width=\maxwidth]{#1}}

\setlength{\parindent}{0pt}

% begin preamble.tex   
\usepackage{calc}
\newlength\vimgflt
\newlength\vposflt
\newlength\vtypesetflt
\newlength\vspaceflt

\usepackage{floatflt}
\renewenvironment{description}{%
    \renewcommand{\includegraphics}[1]{\vspace{-1ex}\Oldincludegraphics[width=4cm]{##1}}
    \renewcommand{\item}[1][]{
        \settototalheight\vimgflt{##1}%
        \global\vimgflt=\vimgflt%
        ##1%
        \end{floatingfigure}%
        \the\pagetotal%
        \setlength\vposflt\pagetotal%
        \global\vposflt=\vposflt%
        }%
    \clearflt
    \hspace{0pt}%
    \begin{floatingfigure}[l]{4cm}}%
    {}

\newcommand{\clearflt}{
    \par
    \setlength\vtypesetflt{\dimexpr\pagetotal-\vposflt}
    \ifdim \vtypesetflt<\vimgflt \setlength\vspaceflt{\dimexpr\vimgflt-\vtypesetflt+\baselineskip} \else \setlength\vspaceflt{0pt} \fi
    \vspace*{\vspaceflt}
    }

\let\Oldsection\section
\renewcommand{\section}{\clearflt\Oldsection}
% end preamble.tex

\begin{document}
\blindtext[3]
\section{Section title}
\begin{description}
\itemsep1pt\parskip0pt\parsep0pt
\item[\includegraphics{example-image-a}]
Lorem ipsum dolor sit amet, consectetuer adipiscing elit. Etiam lobortis facilisis
sem. Nullam nec mi et neque pharetra sollicitudin.
\end{description}

\section{Section title}
\begin{description}
\itemsep1pt\parskip0pt\parsep0pt
\item[\includegraphics{example-image-b}]
Lorem ipsum dolor sit amet, consectetuer adipiscing elit. Etiam lobortis facilisis
sem. Nullam nec mi et neque pharetra sollicitudin.
\end{description}

\section{Section title}
\begin{description}
\itemsep1pt\parskip0pt\parsep0pt
\item[\includegraphics{example-image-c}]
Lorem ipsum dolor sit amet, consectetuer adipiscing elit. Etiam lobortis facilisis
sem. Nullam nec mi et neque pharetra sollicitudin.
\end{description}

\blindtext[2]
\end{document}

Answer 1

The relevant section from the TeXbook on this topic is the following

\ddanger You can look at the $t$ and $g$ values that are used in page breaking by referring to the \ values ^|\pagetotal|' and ^|\pagegoal|', respectively. You can even change them (but let's hope that you know what you are doing). For example, the command |\pagegoal=500pt| overrides the previously saved value of/ |\vsize|. Besides |\pagetotal|, which represents the accumulated natural height, \TeX\ maintains the quantities ^|\pagestretch|, ^|\pagefilstretch|, ^|\pagefillstretch|, ^|\pagefilllstretch|, ^|\pageshrink|, and ^|\pagedepth|. When the current page contains no boxes, |\pagetotal| and its relatives are zero and |\pagegoal| is $16383.99998\pt$ (\TeX's largest \); changing their values has no effect at such times. The integer $q$ in the formula for page costs is also available for inspection and change; it is called ^|\insertpenalties|.

Whenever TeX comes to a possible breakpoint in the current galley it updates the value for "pagetotal" and other parameters and then carries on, unless this breakpoint is making the current page too full. Only then, TeX decides which of the breakpoints on the page is the brest one to actually break the page (it is definitely not the one that triggered the process but it may not be the one before either). So al you see when looking at \pagetotalis how far TeX got before making the page breaking decision.

In your case it actually got to the point where the page got overfull (and that is what you got recorded) and the final break happened earlier. If you use \tracingpages=1 you can see this behavior in the log file, here are the relevant lines from your MWE with the tracing enabled:

% t=346.0 g=550.0 b=10000 p=0 c=100000#
% t=358.0 g=550.0 b=10000 p=150 c=100000#
% t=371.94444 g=550.0 b=10000 p=-300 c=100000#
% t=424.97215 plus 7.16663 minus 0.86108 g=550.0 b=10000 p=0 c=100000#
% t=436.97215 plus 8.16663 minus 0.86108 g=550.0 b=10000 p=150 c=100000#
% t=448.97215 plus 8.16663 minus 0.86108 g=550.0 b=10000 p=250 c=100000#
% t=524.2731 plus 8.16663 minus 0.86108 g=550.0 b=3118 p=-300 c=2818#
<example-image-b.pdf, id=2, 321.2pt x 240.9pt> <use example-image-b.pdf>
<use example-image-b.pdf>
% t=577.30081 plus 15.33325 minus 1.72217 g=550.0 b=* p=0 c=*
[1{c:/texlive/2013/texmf-var/fonts/map/pdftex/updmap/pdftex.map} 

As you can see the last t= on the first page is exactly the one next to your figure B and that one akes the page overfull and so TeX backs up and uses the line

% t=524.2731 plus 8.16663 minus 0.86108 g=550.0 b=3118 p=-300 c=2818# 

as the best breakpoint for the first page.

In summary, \pagetotal helps you understanding where TeX is in its process of breaking the page, but it may show a place beyond the breakpoint that will finally be chosen due to the total costs associated with it. So it is no good for what you try to achieve.