Problems to put off until all (or most) sys_vgui calls are eliminated:
1) gui-side parser inside -- pdgui.js.  Currently we're splitting on newlines so we can separate
   gui_vmess from sys_vgui calls.  This makes it very difficult to handle multi-line msg and text
   boxes.
2) Semicolons -- currently the parser can't tell the difference between semicolons inside symbols
   and semicolons that end statements.  Again, this will be easy to solve once we eliminate sys_gui.

HTML5 which may not be standard yet:
1) mouse.pageX/pageY -- exist in Chromium but maybe not in FF et al.
2) svg 'overflow' attribute -- probably doesn't work in other browsers
3) document.body.scrollTop (might be Chromium-specific, not sure)
3) document.body.scrollLeft (might be Chromium-specific, not sure)

Node-webkit stuff:
1) popup API
2) new window API
3) window menus
4) (probably) present working directory
5) command line argv

Everything else:
* packaging as app, setting correct appname, etc
* get -unique to work (relied on tcl [send] command)
* check if patch windows with screenposition (0,0) get stuck underneath Ubuntu/OSX menu.  If so,
  Node-webkit has a "screen" interface to retrieve the "workable" area of the screen
* choosing the same directory multiple times (see dialog API page)
* figure out why there is a "pd_opendir" global var
* pass k12 mode arg
* implement recent files (using Pd's prefs loading mechanism instead of reimplementing gui prefs)
* implement verifyquit
* save and saveas dialogs aren't defaulting to the present working directory
* create a loop (maybe every second) to check bbox of patchsvg, and set the height/width to that
  bbox for the svg.  That should trigger the correct scrollbar behavior.
* change canvas string to a single int (and investigate any side-effects)
* destroy (or, more likely, hide) selection rectangle when not needed
* change gui_text_select to gui_gobj_select
* make it possible to have the inner cells of hradio change color with the selection, as the border currently does.  Tk just uses the non-hierarchical tags and appends the word "BASE".  (Maybe use an inner <g> to do this.)
* make mycnv use its own selection logic
* abstract away appendChild
* font-size should be set as css property for a class of text, rather than per each gobj in svg
* remove the "fudge_factor" kludge in g_rtext.c, and handle fonts in a sane manner.  In doing so
  we must decide whether or not we want to be able to fetch the bounding box of objects from
  within a Pd patch.  (Currently [canvasinfo] and [objectinfo] are the only objects that can
  do this, though there might also be)
* replace things like gui_select_color with css
* find a better approach to escaping characters for the gui than escape_double_quotes kludge (and handle any other problematic chars)
* look into using <def> and <use> for scalars.
* use Atom's selectors for standard OSX behavior (cut, paste, about, etc.)
* in draw_vis, make circle and ellipse radii attr floats instead of ints
* should stroke-dasharray values be floats?
* once garrays are drawn inside a <g> we won't have to individually erase each child of a scalar
* clean up svg_sendupdate -- put a single gui_vmess at the end
* svg_sendupdate -- make sure there aren't arbitrary %d's that should actually be %g's
* svg_sendupdate -- need a gui interface for stuff like stroke-dasharray (and garrays)
* ibid., for path data, points
* make pdgui function names more consistent
* in draw_vis, move tags to front of function call so they don't have to be an ugly array
* in draw_vis, remove all the coords logic and let svg_togui do that work.  In svg_new,
  leverage the "d", "points", etc. methods to do the same.
* remove the draw_vis warning/code about needing 2 points to draw
* might want to clean up dead windows inside patchwin object (on close)
* make gui_menu_close filename less hacky (currently includes patch args in the middle) 
* figure out why -nrt flag causes Pd to bail on the GUI (it wasn't doing this
  on the amd_64 virtual Jessie machine)
* make dialogs with [yes/no] instead of [cancel/ok]
* make a gui_mess interface for the edge cases that don't require args (similar to sys_gui)
* make sure we're breaking down the socket properly from the GUI side on quit
* nwworkingdir isn't working with the saveas dialog in pdcanvas.html
* nwworkingdir only works on the second file open
* implement the stuff at the end of saveas_callback (recentfiles et al);
* make an "export" menu command for max patch saving.  That way we can accept a single extension
  for saving (.pd), and the native save dialog will warn us on potential file overwrites (even when
  the user neglects to type an extension).
* draw graph inside gobj group
* merge gui_text_select and gui_text_deselect
* in function gui_cord_inspector_update, use parameter "moved" to flash text
* cord_inspector_flash doesn't seem to be deterministic-- if you move among several cords very fast
  it will try to flash a non-existent svg item.  (Probably want to either make a permanent item in
  the pdcanvas.html for it, or hard-code a separate div.)
* for data structures, change "" to 'none'