Revisions to pane_refresh

Instead of passing a pointer to a struct which is then copied and
otherwise (largely) unused, just pass the struct by-value.

Also change the setting of some fields that isn't needed as they
cannot have been changed.

And never use a ci again after it has been passed to a command.

Signed-off-by: NeilBrown <neil@brown.name>

remove call_str and call_extra

no longer used.

Signed-off-by: NeilBrown <neil@brown.name>

Make sure python compares/hashes panes properly.

Two python objects for the same pane must compare and hash the same.

Signed-off-by: NeilBrown <neil@brown.name>

Add 'Comm' type to python interface.

This is a command - it can be called from python to run non-python
code.

There are issues with argument passing that I haven't fully resolved...

Signed-off-by: NeilBrown <neil@brown.name>

Make it possible to replace current event handler.

Setting DAMAGED_EVENTS on the root and then refreshing
should do it.  Untested though.

Signed-off-by: NeilBrown <neil@brown.name>

First step to plugable event loops.

attach libevent via an plugin command.

Signed-off-by: NeilBrown <neil@brown.name>

Use new notifiers for lib-view

Signed-off-by: NeilBrown <neil@brown.name>

use new notifiers to notify changes to render-lines.

Signed-off-by: NeilBrown <neil@brown.name>

Use new notifiers for render-hex.

This makes the he_data redundant.  I'll leave it there as I'll
probably find a need for it later.

Signed-off-by: NeilBrown <neil@brown.name>

Use DEF_LOOKUP_CMD in render-hex.

Signed-off-by: NeilBrown <neil@brown.name>

More fixed for closing popup windows.

I think I need refcounting on panes :-(

Signed-off-by: NeilBrown <neil@brown.name>

popup: Close doc more directly.

I want this to all go eventually with auto-close documents.
But destroying a doc via some pane is problematic as the
pane might have disappeared already.  So do it more directly.

Signed-off-by: NeilBrown <neil@brown.name>

Disconnect pane where possible before closing children.

This limits multiple closes due to recursion to some extent.

Signed-off-by: NeilBrown <neil@brown.name>

Introduce DAMAGED_CLOSED

This gets set when a pane is being closed, and prevents repeated
close when a child tried to close a parent.

Signed-off-by: NeilBrown <neil@brown.name>

Use new notifiers to close a popup when target disappears.

Signed-off-by: NeilBrown <neil@brown.name>

Use new notifications to support search_again

That than the old doc notifer.

Signed-off-by: NeilBrown <neil@brown.name>

Start changing rules about ->focus and ->home

I'm still thinking this through....

'home' must always be the pane where the command was found.
'focus' should be where the action happens.

Previously 'home' would always end up being a parent of 'focus' but
sometimes, particularly calls from popup to target, we want something else.

So allow the caller to set 'focus' to themselves and 'home' to where
to start looking for the command.
If either is home is not set, focuses it moved out to a leaf first.

This still seems a bit clumsy...

Signed-off-by: NeilBrown <neil@brown.name>

Use new notifiers to send notification when document changes.

Signed-off-by: NeilBrown <neil@brown.name>

Unify doc_notify_change and point_notify_change more.

In particular: export only one interface.

Signed-off-by: NeilBrown <neil@brown.name>

Use new notifier to close all views on a document when destroyed.

Signed-off-by: NeilBrown <neil@brown.name>

Add infrastructure to allow panes to get notifications from other panes.

Signed-off-by: NeilBrown <neil@brown.name>

Make struct cmd_info mostly read-only

I want this to be treated readonly except for a few
places where explicit write access is requested.
Partly because it used to be changed a lot and I want to stop that.

Signed-off-by: NeilBrown <neil@brown.name>

Remove doc_from_pane

No more callers.

Signed-off-by: NeilBrown <neil@brown.name>

Change line-count to use a pane rather than a doc.

Now all those pane-based functions are available...

Signed-off-by: NeilBrown <neil@brown.name>

Introduce vmark_new to avoid direct acess to doc_new_mark

Signed-off-by: NeilBrown <neil@brown.name>

Define doc:set-attr and use in line-count.

Signed-off-by: NeilBrown <neil@brown.name>

Introduce pane_mark_attr.

This searches panes for an attribute at a mark.
This removes another doc_from_pane call.

Signed-off-by: NeilBrown <neil@brown.name>

search: used Move-File rather than mark_reset to go to start.

this removes another doc_from_pane call.

Signed-off-by: NeilBrown <neil@brown.name>

Use attributes to get name of document instead of direct access.

This removes another doc_from_pane() call.

Signed-off-by: NeilBrown <neil@brown.name>

Use pane_attr_get to get linecount attributes for 'view'.

This reduced need for 'doc'.

Signed-off-by: NeilBrown <neil@brown.name>

Introduce doc_following_pane and change many mark_next/mark_prev to _pane

still tring to get 'doc' references out of non-core/doc code.

Signed-off-by: NeilBrown <neil@brown.name>

All ed->commands to be accessed by key_handle

This makes sense and means we don't need to find the 'ed'.

Signed-off-by: NeilBrown <neil@brown.name>

Introduce mark_next_pane and mark_prev_pane

Use these to remove the need for 'doc_from_pane' from search_add.

Signed-off-by: NeilBrown <neil@brown.name>

Rationalise prev/next mark function.

- don't need the 'safe' version any more
- don't need to pass 'doc'.
- consistent names: doc_{prev,next}_mark_{all,view}

Signed-off-by: NeilBrown <neil@brown.name>

Change hlist to make the head directlry recognisable.

Store a least-significant-bit in the pointer in the 'head'
so that we can recognise the start of the list without carrying
around the pointer to the head.

This allows free movement around marks without holding the doc.

Signed-off-by: NeilBrown <neil@brown.name>

Remove 'doc' arg from next_mark.

It doesn't need it.

Signed-off-by: NeilBrown <neil@brown.name>

Remove mark_same calls from non-core/doc code.

These should use mark_same_pane.

Signed-off-by: NeilBrown <neil@brown.name>

Remove 'd' from render_complete_line

It isn't used.

Signed-off-by: NeilBrown <neil@brown.name>

Change doc_destroy to take a pane.

This starts to reduce the number of doc_from_pane() callers.

Signed-off-by: NeilBrown <neil@brown.name>

Python: return number from 'call'.

Return value is now more meaningful, so need to pass it back.

Probably should do something better with negative values.

Signed-off-by: NeilBrown <neil@brown.name>

python_call: don't accept return values for storage in 'ci'.

That approach is no longer supported.

Signed-off-by: NeilBrown <neil@brown.name>

Use callback to return marks from doc:vmark-get

Signed-off-by: NeilBrown <neil@brown.name>

Use callback to return pane for call_pane()

Signed-off-by: NeilBrown <neil@brown.name>

Use a call-back to return pane from "display-ncurses".

Signed-off-by: NeilBrown <neil@brown.name>

Use callbacks to return pane from "doc-*"

Signed-off-by: NeilBrown <neil@brown.name>

Use callbacks to return string from Complete:prefix.

Also fix up freeing of strings.

Signed-off-by: NeilBrown <neil@brown.name>

Mark the '42' hack in render-complete as such.

And hide it from future attempts to make 'ci' read-only.

Signed-off-by: NeilBrown <neil@brown.name>

Return value from doc:add-view directly

Just like doc:find-view

Signed-off-by: NeilBrown <neil@brown.name>

Remove assignment to ci->mark in doc_file.

It isn't needed, and I want to make ci read-only.

Signed-off-by: NeilBrown <neil@brown.name>

complete: remove the finding of the render-lines pane.

If we just make render-lines use "key_handle_focus", it is OK
for render-complete to be placed on top of the stack.
This makes things a lot cleaner.

We now need to let move-line through too.

Signed-off-by: NeilBrown <neil@brown.name>

Change "render-line" to return line via a callback.

Signed-off-by: NeilBrown <neil@brown.name>

render-lines: don't set ci->hy unnecessarily.

I want 'ci' to be read-only inside commands.

Signed-off-by: NeilBrown <neil@brown.name>

Use REDEF_CMD to predeclate keymap_attach_func.

Signed-off-by: NeilBrown <neil@brown.name>

Use callbacks to return pane from render-*-attach commands.

Signed-off-by: NeilBrown <neil@brown.name>

use callbacks to return a pane from 'attach' commands.

Signed-off-by: NeilBrown <neil@brown.name>

Use callback to return string from doc:get-str

This way it is clear how ownership of the string is managed.

Signed-off-by: NeilBrown <neil@brown.name>

Use a callback to return string from doc:get-attr

Signed-off-by: NeilBrown <neil@brown.name>

Use a callback to return mark for mark_at_point.

This might seem a bit forced, but I really think callbacks
like this are the best generic approach.  I hope.

Signed-off-by: NeilBrown <neil@brown.name>

Unify calls to doc:dup-point.

All these calls can use mark_to_point - so do that.

Signed-off-by: NeilBrown <neil@brown.name>

Change Complete:prefix to return count directly.

with an offset of '1' to avoid returning '0'.

Signed-off-by: NeilBrown <neil@brown.name>

Change text-search to return result directly.

-2 is cannot find
-1 is incomplete regex
0 is no such commnd
N is match of length N-1 found.

Signed-off-by: NeilBrown <neil@brown.name>

Change doc:replace to return result directly.

There isn't must result here - just a flag to say "yes, a change was made".
But it was being handled completely wrongly!
I think it is fixed now.

Signed-off-by: NeilBrown <neil@brown.name>

Change doc:mark-same to return result directlry.

'1' means they are the same.
any other +ve number means they are different.

Signed-off-by: NeilBrown <neil@brown.name>

Change doc:step to return character directly.

Don't use ci for returing things.

As a char could be zero and WEOF could appear negative,
use high bits to make the number look positive.

Signed-off-by: NeilBrown <neil@brown.name>

Change doc:find-view to return value directly.

I'm moving away from using the cmd_info to return stuff.
So for numbers, just return them directly.

Signed-off-by: NeilBrown <neil@brown.name>

Update README file

Lots has happened since this was written...

Signed-off-by: NeilBrown <neil@brown.name>

python: make standard globals available to python scripts.

Also make 'editor' and 'edlib' available to subsequent function calls

Signed-off-by: NeilBrown <neil@brown.name>

Make symbols in loaded libraries global.

I don't really like doing this, but python needs it for
"import gtk" to work.

Signed-off-by: NeilBrown <neil@brown.name>

Fix various issues with python

Now that I've started testing a bit more...

Signed-off-by: NeilBrown <neil@brown.name>

Remove 'buf' from pane_text

It isn't used.

Signed-off-by: NeilBrown <neil@brown.name>

python: Add 'init' function for panes.

This calls pane_register.

Signed-off-by: NeilBrown <neil@brown.name>

Python: add interface for pane_focus and pane_refresh.

Signed-off-by: NeilBrown <neil@brown.name>

Create a command for defining new named commands.

Signed-off-by: NeilBrown <neil@brown.name>

Initialise doc_default_cmd earlier.

It needs to be initialised before the very first doc is created.

Signed-off-by: NeilBrown <neil@brown.name>

pane_refresh: always refresh whole display.

Anything else doesn't make sense.

Signed-off-by: NeilBrown <neil@brown.name>

pane_refresh: allow post-order "Refresh" calls, by request.

This allows the display to flush out updates at the end,
without needing the '0-timeout' hack.

Signed-off-by: NeilBrown <neil@brown.name>

Discard concept of persistent key modifiers.

I'm not using them and I no longer think they make sense.
The goal they were meant to serve is better served by
setting a new "global-set-keymap".

Signed-off-by: NeilBrown <neil@brown.name>

separate keyboard-input state into separate module.

This localises the concept and makes it possible to access
the state by sending commands.

Signed-off-by: NeilBrown <neil@brown.name>

Fix search bug: not finding longest match.

We need to keep going while we have a partial match (len=-1)
and record the last good match.

Signed-off-by: NeilBrown <neil@brown.name>

search: fix but with dot matching newlines.

Also default to case-insensitive until I get that configurable.

Signed-off-by: NeilBrown <neil@brown.name>

Search: note in prefix when search has wrapped.

Signed-off-by: NeilBrown <neil@brown.name>

use ->home->data, not ->focus->data

->focus isn't guaranteed to be the local pane.  Home is.

Signed-off-by: NeilBrown <neil@brown.name>

Search: move to 'search start' when finishing the search.

This ensures the final search is from the current starting point.

Signed-off-by: NeilBrown <neil@brown.name>

Better handling of formatting with render-complete.

We highlight the common prefix and stripe all highlights
and quoting before using the string.

Signed-off-by: NeilBrown <neil@brown.name>

Fix bug causing render-complete to show a non-matching first line.

That "Move-to" needs to get through.

Signed-off-by: NeilBrown <neil@brown.name>

search: make sure point is visible when search is aborted.

Signed-off-by: NeilBrown <neil@brown.name>

render-format: allow '<' characters in format fields.

These need to be quoted.

Signed-off-by: NeilBrown <neil@brown.name>

Use the © copyright character is copyright messages.

Because I can...

Signed-off-by: NeilBrown <neil@brown.name>

Add lib-messageline.

This places a line on the bottom on the display and displayed messages there.

Signed-off-by: NeilBrown <neil@brown.name>

ncurses: allow use of full screen.

If we want to trim a line off the bottom, do it separately.

Signed-off-by: NeilBrown <neil@brown.name>

Save search string for later re-use.

Signed-off-by: NeilBrown <neil@brown.name>

Disable recursive popups except by permission.

So trying to search in a file-file buffer now doesn't work.

This might need work for searching in a completion list??

Signed-off-by: NeilBrown <neil@brown.name>

tile: fix 'next' to walk through all pop-ups.

Still need to work on 'prev'!

Signed-off-by: NeilBrown <neil@brown.name>

Move root popups to the tile-root.

Having popups always attached to tiles seems clean and will make
pane-movement easier.

Signed-off-by: NeilBrown <neil@brown.name>

Page up/down now move cursor to EOF when appropriate.

emacs doesn't do this, but I think it looks better... maybe.

Signed-off-by: NeilBrown <neil@brown.name>

Fix page-up/down in directory listings.

If the target for cursorXY is in the header line, we need to adjust.

Signed-off-by: NeilBrown <neil@brown.name>

Introduce some 'call*' functions to call command.

This simplifies lots of code.

Signed-off-by: NeilBrown <neil@brown.name>

Quote regex chars when they are included from doc into search string.

Also handle unicode search properly.

Signed-off-by: NeilBrown <neil@brown.name>

text: fix bug with adding at end of allocation.

Signed-off-by: NeilBrown <neil@brown.name>

Rename comm_ncurses to display_ncurses

"comm_" is part of an old deprecated naming scheme.

Signed-off-by: NeilBrown <neil@brown.name>