path: root/todo

-*- mode:outline -*-

$Id$

* Priorities
============
A* (SUPERSONIC URGENT)  to be fixed yesterday.
A (URGENT) to be fixed immediately for pre-release 
B (High)   to be fixed before release (Version 2.0)
C          would be nice to fix before release of 2.0; but not crucial
D (Medium) desirable to fix at some point
X (Low)    probably not worth wasting time on

* This is a list of things which need doing in the generic interface
====================================================================

A* multiple files bug fix:
   It can happen (in Isabelle) that the prover retracts a file
   which asks for another to be retracted which is *not* on
   proof-included-files-list.  The same case could perhaps 
   happen in lego if we relax the restriction on switching
   scripting buffer?  We need to remove the locked regions
   from *any* buffer which is matched by the retract_files_regexp,
   not just those in proof-shell-compute-new-files-list.
   --> One case now fixed by allowing current scripting buffer to
    be retracted.  Are there other cases?  (da to investigate)

A Revise ProofGeneral.texi and publish LaTeX version as an LFCS
  Technical Report (2+2 days; da + tms)

A Check that byte compilation (and compiled code!) 
  works for both varieties of Emacs. 
  Fill out Makefile.dist (2hr da)

B Make completion more generic. For Isabelle and Lego, we can build a
  completion table by querying the process, which is better than
  messing with tags.  
A* fix any bits I've broken  (da, 1hr)

A* Fixup for non-script buffer locking:

    proof-locked-end called from wrong buffer when error message
    is output.

    proof-restart-script is now broken (2h da)

B Add proof-quit-command: some provers may like a quit command to be
  sent to the shell, not just EOF !  (see proof-stop-shell).
  Also reconcile proof-restart-script and proof-stop-shell, see
  comments in code. (1h da)

B da: goal-hyp: this should be more generic.  At the moment, there are
  generic variables which are only used in the specific code:
  proof-shell-goal-regexp and proof-shell-assumption-regexp.
  This is confusing.  I suggest making lego-goal-hyp the
  default behaviour for proof-goal-hyp-fn a hook function.
  That will work for Isabelle too.   (15 mins)

C  proof-goal-command-regexp: add this setting to coq.el.
   Rationalize use of proof-goal-command-p
   (probably can be scrapped now).  

C  Strange behaviour when killing and re-visiting files that
   haven't been completely processed.  Since they stay on
   the proof-included-files-list, on re-visiting, they are
   marked atomic as completely processed!  
   Possible fix:  if file hasn't been completely processed, 
   maybe retract (and hence remove from proof-included-files-list),
   or query the user?

C  The semantics of `proof-script-buffer-list' is ambigous. The first
   element is the current script buffer, we are not quite sure what
   role the other elements ought to have. We propose to abandon
   `proof-script-buffer-list' and only keep a record of the current
   proof script buffer in `proof-script-buffer'. We can easily check
   dynamically if a buffer is instrumented for scripting, contains a
   locked region, etc. (4h, including testing)

C Improve indentation code and see why it's so slow (at
  least for Isabelle).   Enable it for particular provers if
  it works okay  (but must test in on large files).

C Regions in script buffer have nice "name" property and configurers
  have to set regexps carefully so that it works, but is it actually
  used anywhere?   What's it good for?

C ROBUSTness: deal gracefully with possibility that goals buffer is
  killed during session.   (2h)

D Add support to filter out unwanted noise from the prover by setting
  up a regular expression  proof-shell-noise-regexp (1h)

C support for templates e.g., in LEGO it would be nice if, by default, 
  fresh buffers corrsponding to file foo.l would automatically insert
  "Module foo;" (1h)

C  Remove "FIXME notes" which are just notes I've put in about old
   code in case something breaks   (da, 10mins).

C  Check on all FIXME notes.

C  automise testing procedures in etc/

C  Write proof-define-derived-mode which automatically adds a 
   call back hook somehow.  Propose this to maintainer of derived.el.
   (1.5hrs)

C  Toolbar icon for `proof-execute-minibuffer-cmd'  (1.5hrs)

C Improve web pages after suggestions in doc/notes.txt
  (da, 1.5hrs)

C proof-issue-goal should refuse to work when a proof is in progress.
  Similarly proof-issue-save should refuse to work when a proof
  hasn't been completed! Algorithm: Search the last goal and check
  length of span.  (45min)

C Bug in proof-retract-until-point when called twice in succession:
  calls backward-char at beginning of buffer.  (Should say no
  locked region, or "nothing to retract").  Problems is that
  there is a proof-locked-end, but it's at (point-min). (30min)

C Clean up proof-assert-until-point behaviour.  Discussion results:
    1.  At the moment we get an odd error if it is run in
    the locked region.    This is a bug.  Should behave same
    as proof-assert-next-command in this case (simpler) or
    give better error message.

    2.  At or before first character of proof command, "nothing to do"
    error message.  This is a bug.  Should behave same
    as proof-assert-next-command in this case.

    [1,2 have been "fixed" by rebinding C-c RET to 
     proof-assert-next-command for now, eventually 
     this may form a new version of proof-assert-until-point]

    3.  Movement and possible insertion of spaces/newlines for 
    when new commands are added to the buffer needs careful
    thought! (1h)

C A less drastic version of proof-restart-script would be useful:
  one that doesn't involve killing off the proof assistant process
  and restarting that -- it can take ages! 
  
  We want two different modes of restarting:

     1. Restart all: clear all context and kill proof process.
     2. Restart some: clear context, do some retraction/kill goals
        in proof process.

  Add kill buffer hook to script buffer which does restart some
  in case it's the active buffer.
  (da, 20mins)
  
C Improve proof-goal-command and proof-save-command:
  `proof-goal-command' should be more flexible and support a
  placeholder for the name and the actual goal. In LEGO, we have
    Goal foo : absurd;
    ...
    Save foo;
  Perhaps functions at the generic level to make suitable
  values for the hook, e.g.,

     (setq proof-goal-command  (proof-prompt-named-goal "Goal %s :" "%s;"))

C Cleanup handling of proof-terminal-string.  At the moment some
  commands need to have the terminal string, some don't.
  da: I suggest removing proof-terminal-string altogether and
  adding back the semi-colons or fullstops at the specific level.
  It's not a big deal and would support other provers which
  may use a mixture of terminators, or no terminators at all.

C Better support for adding a new prover: give error messages which
  hint at what variable to set (see proof-issue-goal for example).

C Functions for next,previous input a la shell mode, but in proof
  script buffer (3h, da).

C User-level functions: 
  1. add new version of undo-until-point which behaves analogously to 
  proof-assert-next-command.
  2. make version of proof-restart-script which will start or
  restart the proof assistant as appropriate.  (It's handy to have
  a direct function to start the proof assistant).
  (1hr, da)

C Write test schedule for things to try out with a new instantiation
  of Proof General.

C Add skeleton instantiation files for a dummy prover "myassistant" to
  make it easier to add support for new assistants -- looking at
  any of the existing modes is confusing because of the
  prover-specific stuff.  Ideally it should work for one of
  the default provers as an impoverished example mode.
  (2h, da will do for Isabelle)

C file handling could be more robust; perhaps one should always cd to
  the directory corresponding to the script buffer (currently only
  done for the buffer which starts up the proof system). This could be 
  achieved with a hook which is not set by default. [Remember to add
  user documentation] (30min)

C Reengineer *-count-undos and *-find-and-forget at generic level
  (3h)

C Unify toolbar and menu functions. (1h)

D Multiple files: handle failures in reading ancestors.

D Provide a sensible default frame/buffer layout (4h)

D Implement mouse drag-and-drop support for selecting subterms in the
  goals buffer and copying them in a script buffer. This could be
  implemented by defining button2 in the response buffer and setting
  button2up in script buffers. (1h)

D Add support to proof.el for *not* setting variables for
  commands which aren't supported by a prover.  For example,
  in Isabelle there is no such thing as killing a goal.
  For the minimum set of variables to cover, see FIXME's in isa.el
  (da, 1.5hrs)						   

D proof-find-next-terminator is too slow when it needs to parse
  a long buffer.   Generally a performance problem with
  proof-segment-up-to.
  				    
D Implement proof-find-previous-terminator and bind it to C-c C-a
  (45min)

D Allow bib-cite style clicking on Load/Import commands to go to file.

D Remove duplication of variables e.g., proof-prog-name and
  lego-prog-name for Coq and Lego. (1h)

X  Customization mechanism: is there a way to make saved settings
   not be overwritten by setq's in the code?  Need to think how
   to do this, something like customize-set-variable-if-unchanged
   Otherwise no so useful for people to use customize for
   prover config settings (we'd need to shadow them all again for
   each assistant for this to work smoothly).
   Sure sign saving will fail is when you see "this option has
   been changed outside customization buffer"
   At the moment even setting config variables in a hook is tricky,
   because proof-config-done is called before the hook variables for the
   new mode are.  Our new version of define-derived-mode needs
   to address this.

X Proofs in Isabelle scripts can be non-interactive.  Non-interactive
  proofs have only one command, effectively.  It might be useful
  to find a way to integrate these into Proof General nicely.

X Code in proof.el assumes all characters with top bit set are special
  instructions to Emacs.  This is rather arrogant, and precludes proof
  systems which utilize 8-bit character sets!  Needs repair.  (3h)

X Span convenience functions for special 'type property.
  Could put these in proof.el or somewhere.

  (defsubst set-span-type-property (span val)
    "Set type property of SPAN to VAL"
    (set-span-property span 'type val))

  (defsubst span-type-property (span)
    "Get type property of SPAN"
    (span-property span 'type))

  etc.   (1hr)

X Read-only mode of extents sometimes gets in the way: for example,
  if file changes on disk, can't reload it via usual functions.
  Can this be improved?  Always have to retract first, and that
  always leaves stuff around.

X toolbar icons: Automatically generate reduced and
  pressed/greyed-out versions from gimp xcf files.  Keep the
  xcf files under CVS rather than xpm files.  (2h, da)

X Remove .gif, .jpg, .xpm, .xbm files from the CVS repository.  Add 
  .cvsignore's for them instead.  Disadvantage: developers will need
  to have the Gimp installed to build them via 'make images'
  (or copy them from the latest download).   (da, 1hr)

X proof-site (da): I think it would be nice to change the architecture
  to make customization for new provers much easier.
  The standard use of 'define-derived-mode' could be invoked
  automatically in proof-site, and we could easily get away from the 
  kludge of proof-config-done and friends.
  (Compare this to the way font-lock works automatically for XEmacs
  when the right variable name is set, but for FSF Emacs you have
  to write something special).
  The objection to doing this is based on the idea that we should use
  an object-like inheritance mechanism to define the new modes.
  But if this is forgone, it might even be possible to add
  support for new assistants entirely via the customize mechanism,
  without any knowledge of elisp apart from regular expressions!

X Support a history of proof commands, with a "redo" command to
  redo undo-to-point or sequences of toolbar undo's.

X Support for x-symbols package.
  Provers with sophisticated/configurable syntax should tell Emacs
  about their syntax somehow, rather than trying to duplicate
  specifications inside Emacs.

X Comment support is not very generic: we don't support end-of-line
  terminated comments.  Is there any case where this might be
  worthwhile?  (2h to add it).

X Write a Makefile for the distribution.  It can do things like
  install the info file properly.  The work is at the moment done
  in the RPM spec file instead.

X Make process handling smarter: because Emacs is single-threaded,
  no process output can be dealt with when we are running some
  command.  This means that it would be safe to extend the 
  red region, by putting more commands on the queue.  Also it would
  be safe to implement a clever undo command which worked on the
  red region: if there are commands waiting to be processed, we
  could remove them from the queue.  If there are no commands waiting,
  we have to wait until something becomes blue to undo it by sending
  a command to the process.

X Ideas for efficiency improvements.  Rather than repeatedly
  re-parsing the buffer, we could parse the whole buffer *once*
  and make adjustments after edits, like font-lock.  We could
  make an extent for every command, and set it to "blue", "red"
  or "clear" as appropriate.  (This would allow proofs to be
  sent out-of-order to the proof process too, although perhaps
  that's not so nice).
  The function proof-segment-up-to could be made to cache its
  result.

X We need to go over to piped communication rather than ptys to fix
  the (Solaris) ^G bug. In this circumstance there's a bug in the
  eager annotation code. Document this problem so that it can be
  tested for future versions. [Currently the problem is documented in
  Email messages sent to lego]

X proof-mark-buffer-atomic marks the buffer as only containing
  comments if the first ACS is a goal-save span. This is however not a 
  problem for LEGO and Isabelle. (30 min)

X Idea for future re-engineering:
   Indirect Buffers seem to be a cunning way
   to implement the response buffer and goals buffer, since they're
   basically variants on displaying fragments of the shell buffer
   output. Unfortunately seems to be implemented only in FSFmacs at the
   moment.


* Proof-by-Pointing
===================

B Change proof by pointing (pbp) stuff into proofstate buffer stuff.
  (1h tms)

C Fixing up errors caused by pbp-generated commands; currently, script
  management unwisely assumes that pbp commands cannot fail (2h)

B Rename pbp-mode to response-mode or goals-mode (which doesn't
  support any actual proof-by-pointing) (30min) 

B Outsource actual pbp/goals functionality (30min tmd)
  (separate pbp annotations from other annotations).
  Make a file proof-goals.el.

X pbp code doesn't quite accord with the tech report; in particular it
  decodes annotations eagerly. Lazily would be faster, and it's what
  the tech report claims
  --- djs: probably not much faster, actually.


* Here are things to be done to Lego mode
=========================================

C In LEGO, apart from Goal...Save pairs, we have
  declaration...discharge pairs. See the section "Granularity of
  Atomic Commands" for a proposal on how to generalise the current
  implementation so that it can also deal with "Discharge".
  [See also the Coq problem with Sections] (6h)

C Inoking an "Expand" command produces a new proof state. But this is
  incorrectly displayed in the response buffer and not the goals
  buffer because special annotations are missing. Presumably, one
  ought to replace "Toplevel.Pr()" by "Toplevel.PR()" in the
  definition of "Expand" (see file newtop.sml [Version 1.4]). (30min)

C fix Pbp implementation (10h)

B release new version of the LEGO proof engine (4h tms)

X Mechanism to save object file

D Improve legotags. It cannot handle lists e.g., with 	

    [x,y:nat];

  it only tags x but not y. [The same problem exists for coqtags]

* Here are things to be done to Coq mode
========================================

C Retraction in a Section should retract to the beginning of the whole 
  section. See the section "Granularity of
  Atomic Commands" for a proposal on how to generalise the current
  implementation so that it can also deal with sections.
  [See also the LEGO problem with Discharge] (6h)

A* Implement multiple file handling.

C derive a coq-response-mode from proof-response-mode; see lego.el (10min)


D set proof-commands-regexp to support indentation for commands
						    (10min)

D Add Patrick Loiseleur's commands to search for vernac or ml files.

X Sections and files are handled incorrectly.

D Proof-by-Pointing (10h)

D Add coq-add-tactic with a tactic name, which adds that tactic to the 
  undoable tactics and to the font-lock. (2h)

D Improve coqtags. It cannot handle lists e.g., with 	

	Parameter x,y:nat

  it only tags x but not y. [The same problem exists for legotags]

* Here are things to be done to Isabelle Mode
=============================================

A* Check all regexps are suitable instantiated.

A Fixup multiple file handling with theory loader hacks to Isabelle.

C derive an isa-response-mode from proof-response-mode; see lego.el (10min) 

C Obscure process-handling problem which prevents goal appearing
  when Isabelle HOL is started.  Visit example.ML, do "next command".
  Suggested process: debug proof-shell-filter.  Has something to
  do with annotated prompts.
  (4h)

D Implement completion for Isabelle using tables generated by
  the running process.

D Add useful specific commands for Isabelle.  Many could
  be added.  Would be better to merge in Isamode's menus.
  (probably a month's work to bring together Isamode and proof.el,
   making some of Isamode generic)

X Add ability to choose logic.  Maybe not necessary: can use default
  set in Isabelle settings nowadays, in the premise that most people
  stick to a particular logic?   But then no support for loading
  user-saved databases.
  (ponder this)

X Write perl scripts to generate TAGS file for ML and thy files.
  (6h, I've completely forgotten perl), or better:

X Manage multiple proofs (markers in possibly different buffers)
	 
* FSF Emacs
===========

C `proof-zap-commas-region' does not work for Emacs 20.2 on
  lego/example.l . On *initially* fontifying the buffer,
  commas are not zapped. However, when entering text, commata are
  zapped correctly. (2h)

* Release
=========

A add a new page/section, maybe a make mechanism (but maybe
  instructions instead) for a stable release.  

B add links *from* Coq, Lego & Isabelle Web pages (da, tms 10 min)