Hi Hans, Taco and Mojca,
There is a bit of inconsistency between the two subformula numbering
methods (\startsubformulas that I wrote, and \NR[+][a] of
\startalign). While writing a paper I realized that both of them are
needed. For example
\startsubformulas[eq:encoder] \placeformula
\startformula \startalign
\NC Z_1 \EQ c_1 (X_1), \NR[eq:encoder 1]
\intertext{and for $t=2,\dots,T$,}
\NC Z_t \EQ c_t (X_1, \dots, X_t, \tilde Y_1, \dots, \tilde Y_{t-1}).
\NR[eq:encoder t]
\stopalign \stopformula
\stopsubformulas
and
\placeformula
\startformula \startalign
\NC \hat X_t \EQ g_1(Y_1) \NR[eq:decoder][a]
\NC M_1 \EQ l_1(Y_1) \NR[eq:memory][a]
\intertext{and for $t=2,\dots,T$,}
\decrementnumber[formula] \decrementnumber[formula]
\NC \hat X_t \EQ g_t(Y_t, M_{t-1}) \NR[+][b]
\NC M_t \EQ l_t(Y_t, M_{t-1}) \NR[+][b]
\stopalign \stopformula
However, the startsubformulas uses \@@fnseparator as separator while
\NR[+][a] uses no separator. It is relatively straight forward to make
\NR honor the separator, in \dododoformulanumber change
\edef\hetnumber{#2} to \edef\hetnumber{\@@fnseparator#2}
and
\edef\hetnumber{#4} to \edef\hetnumber{\@@fnseparator#4}
To maintain backward compatibility we can change
\setupsubformulas[separator=\@@fmseparator]
to
\setupsubformulas[separator=]
Any comments? I think that Mojca and I are the only people who use
\startsubformulas right now (my fault, have not documented them), so
breaking the backward compatibility of \startsubformulas should not be
too big a problem.
Aditya