fped/TODO

Major missing features:
- add postscript output (partially done)
- add option to include/omit helper vecs and frames (done for display, still
  need postscript). Better idea: in PS, print the component 10x, 1x, and then
  each individual frame 10x.

Minor missing features:
- reorder frames (can use text editor)
- reorder rows in a table (can use text editor)
- reorder columns in a table
- reorder variables in a frame (can use text editor)
- move items/vectors up and down the hierarchy

Error detection:
- eliminate duplicate instances

Style and usability:
- make column of entry field greedily consume all unallocated space
- make menu bar consume all unallocated space instead of sharing it evenly with
  upper toolbar
- status area looks awful
- add button with GTK_STOCK_UNDELETE for "undelete" to menu bar
- maximizing pad name size creates uneven text sizes. Particularly "1" gets
  excessively large.
- pango_layout_get_size doesn't seem to consider rotation, so we currently
  don't properly size rotated text.
- when changing the part, we should automatically switch to a configuration
  that generates any of its (non-global) elements
- add zoom controls to top toolbar

Bugs:
- default silk width has no business being hard-coded in obj.c
- undelete only works if not much has changed since the deletion
- focus should return to canvas if nobody else wants it
- whenever we call parse_* for input parsing, we may leak lots of expressions
- can't edit measurement labels through the GUI
- unbalanced parentheses in text throw off Postscript syntax

Code cleanup:
- merge edit_unique with edit_name
- merge find_var_in_frame with similar mechanisms in expr.c and fpd.y
- add regression tests
- the drag logic is too complex. Better: let tool/instance just generate the
  list of points at each stage, then handle the highlighting and hovering
  inside a dragging module.
- code organization is very poor. E.g., functions belonging to the different
  items (pads, silk objects, vectors, etc.) should be grouped by item, not by
  type of function, similar to how some things are now with gui_meas.c
- eval_string_var should be merged into eval_var and the result should be a
  struct num (?) that can contain both types. This also means changing all the
  ops to handle/reject strings.

Open decisions:
- Q: should loop be (start, last) or (start, iterations) ? or start ... last ?
- change vector circle color ? (also, highlight on hover ?)
- Q: allow reassignment of vector names ?
  A1: no: would cause confusion in GUI (vectors could become orphaned)
  A2: yes. but we don't change the linkage.
- Q: how do we handle stacks of objects ?
  A: we don't but we make it easy to avoid them, by giving a good zoom,
     flexible selection, and by disallowing stacks of identical objects in the
     first place.
- Q: add frame arguments ? (e.g., .frame pad(pin_num_offset) ...)
  we can already approximate this by introducing an intermediate table that
  sets up the arguments (provided that we don't consider vectors as well)
- Q: should we make it a requirement to generate objects only once ?
  A: yes.

Future directions:
- future: consider using cairo instead of gdk
- live update of value when entering strings and expressions ?
- advanced: non-standard solder mask
- advanced: solder paste exceptions (subtractive, additive)
- advanced: holes
- advanced: silk line width
- future: consider editing non-canvas items (e.g., variable names/values) in
  place
- add a means to cut&paste and copy&paste objects, and perhaps also variables
  (to move objects, we need to make sure all references are included. besides
  that, copy&paste should be a slight variation of delete/undelete)
- instead of blue screening, we could perhaps just skip the offending items and
  replace them with a "here's a problem" marker that would still point to the
  underlying object and allow repairs to be made
- instead of the awkward lock-switch-anchor process for frame references, we
  could just drag the frame name into the canvas