valgrind complaints when running datauri

Jeremy's valgrind logs have had a couple of these recently, e.g.,
http://starurchin.org/dillo/valgrind/5b601af9ce7facc17690b88c4323c194fded01c9.html

I found that
- after dpidc stop
- valgrind --leak-check=full --trace-children=yes dillo
- going to http://jimbojw.com/wiki/index.php?title=Data_URI_Explained
- and pasting in the second example (the percent-encoded one, not the
  base64 one)
gets me:

==20817== Invalid read of size 4
==20817==    at 0x804AED5: dStr_insert_l (dlib.c:287)
==20817==    by 0x804AFB5: dStr_append_l (dlib.c:308)
==20817==    by 0x8049317: main (datauri.c:332)
==20817==  Address 0x41cbbb0 is 0 bytes after a block of size 664 alloc'd
==20817==    at 0x402AE29: realloc (vg_replace_malloc.c:662)
==20817==    by 0x804A736: dRealloc (dlib.c:53)
==20817==    by 0x804918B: datauri_get_data (datauri.c:273)
==20817==    by 0x80492E1: main (datauri.c:324)
==20817== 
==20817== 
==20817== Process terminating with default action of signal 11 (SIGSEGV): dumping core


Valgrind has options for debugging, but I haven't had any luck with
getting them to work so far.

I'm imagining that this might have to do with only being able to send
part of the data at a time when everything's sluggish due to valgrind,
but I don't know datauri/dpip well enough to go straight for the most
likely spots...

PS I don't trust the "by 0x8049317: main (datauri.c:332)" line because
I have -O0, and sometimes it's had some 0x40... address instead that it
describes as being below main.

PPS I find myself wishing that we could have the dpis back in dillo.