Am 18.08.2018 um 00:26 schrieb Karl Berry:
Martin: please see https://tug.org/pipermail/tex4ht/2018q3/001984.html
and thread following about slow operation handling a huge document (not
surprising in my view).
Is it perhaps possible to cache results so that dvisvgm does not have to
recompute every math fragment every time? Any other possibilities come
to mind?
As far as I can tell, there isn't much you can do to speed up the
conversion process of dvisvgm itself. Every DVI document requires some
time-consuming computations that can't be accelerated easily.
Especially, the conversion of fonts/glyphs to SVG equivalents slow
things down. It gets even more complicated if PostScript code is
involved and/or you have to process thousands of DVI files or pages.
However, technically it shouldn't be necessary to convert all math
fragments every time the document is processed. In the (proprietary)
infrastructure at our university we only convert the portions that have
actually changed. This is simply done by using md5-based file names
computed from the corresponding LaTeX code. Before starting the actual
conversion, the system checks if there's already an SVG file present
which matches the hash value of the LaTeX code. If so, running LaTeX and
dvisvgm can be skipped. This also has the advantage that every fragment
is created only once even if it's referenced multiple times in the document.
Since my knowledge of TeX4ht is pretty limited, I don't know if it's
possible to realize a similar mechanism here.
Regards,
Martin
