Write up systemic issues with mgp2pdf

BUGS.rst

@@ -1,3 +1,5 @@
+Bug examples:
+
 - samples/doctests/doctests.mgp, slide 8: image should be centered
 
 - samples/doctests/doctests.mgp, slide 17: image should be above text
@@ -26,3 +28,42 @@
 - samples/python/python7.mgp: slide 33 is rendered wrong,
   compared to the original PDF from 2006 (same problem as the above
   actually)
+
+Systemic bugs:
+
+- %mark/%again handling is wrong: you can use %again only once
+  (https://github.com/mgedmin/mgp2pdf/pull/1)
+
+- %mark/%again handling is wrong: %again isn't supposed to add a new blank
+  line and apply defaults etc (but I rely on that in
+  samples/pyconlt/talk.mgp!)
+
+- %defaults are applied too late: when I see text.  So if I have
+
+     %left
+     text on line N
+
+  I'm handling %left before I'm applying the defaults of line N, so I have
+  to hack _directives_used_in_this_line to suppress default application,
+  which is totally the wrong mechanism to solve this.
+
+  Also, defaults do not get applied for %newimage, which causes
+  the image alignment bug in samples/doctests/doctests.mgp
+
+- %newimage is supposed to break lines:
+
+    %newimage "a.png"
+    %newimage "b.png"
+
+  mgp puts them on different lines, mgp2pdf puts them on one line.
+
+- %cont is weird, e.g.
+
+    %newimage "a.png"
+    %cont, newimage "b.png"
+    %newimage "c.png"
+
+  puts all three images on a single line in mgp, which I don't understand
+
+- %deffont is handled totally incorrectly; it's just a macro that
+  records a bunch of directives to be replayed when the user uses %font