Albert Gräf authored 3 years ago and Gabriela Bittencourt committed 3 years ago

The autcomplete option globally enables or disables autocompletion.

The autocomplete_prefix option only matches object name prefixes,
instead of looking for the text anywhere in an object name. So, e.g.,
typing "me" lists "metro", but not "timer" as possible completions if
this option is enabled.

Both options are currently disabled by default.

TODO: French translations.

The autocompletion of messages and comments doesn't properly deal with
line breaks in the current implementation. Disable those for now.

This adds a dropdown menu with completions from the completion index
when the user creates a new object, message or comment, and starts
typing. Note that this index is initially populated with objects from
the search index of the help browser. As the user types object names,
arguments, messages and comments, they will be added to the completion
index as well.

Entries from the menu can be chosen with the cursor up and down
keys. The enter key can then be used to select an entry and insert the
corresponding completion. Alternatively, clicking with the mouse also
selects an entry.

NOTES / TODO:

This is a very first implementation of autocompletion for purr-data, and
as such it still has a few minor quirks and shortcomings:

- Help index completions: Completions from the help browser's index will
  only be available once that index has been built. By default, this
  happens when the help browser is first launched. However, you can
  change this so that the help index is automatically created when
  purr-data launches, by ticking the corresponding checkbox in the GUI
  preferences. This will make sure that completions from the help index
  are always available, even before launching the help browser for the
  first time.

- Prefix matches: As shipped, completions will encompass all matches of
  the typed text, even within an index item. There is an option to
  select prefix matches only in the code of the search_obj() function in
  pdgui.js. Currently changing the code is the only way to get this
  behavior; there should probably be a checkbox in the GUI preferences
  to make this easier.

- Tab completion: There is no binding for the Tab key in order to select
  a default completion yet, so currently you have to use the cursor keys
  and enter or the mouse to select a completion. This will hopefully be
  added in the future.

- Stale completions: At present, there is no way to remove "stale"
  completion entries, such as obsolete entries from the help index, or
  manually typed completions which haven't been used for a long
  time. There should probably be a way to do this in a (semi-)automatic
  fashion, but at present the completion engine offers no support for
  this.

For now the results are shown in the purr-data console,
printing the first 10 completions.

- create the completion_index at the initialization of purr data
- add newly created objects to the completion_index
- update the completion_index by adding all objects from the index
  of the help browser whenever the index is created via make_index
- the completion_index is saved into a json file when purr-data
  quits, and also every 50 changes in order to mitigate data loss
  in case purr-data crashes

Pass 1 now iterates over the default hierarchy (either doc or the entire
libdir) and the help path (if enabled) to determine names and locations
of all help patches in the scope of the help browser (as set in the GUI
preferences). This initial pass is needed, in particular, to resolve
related object references in the meta data in pass 2.

Pass 2 then iterates over all index entries constructed in pass 1,
adding all the available meta data, including cross references to
related objects. When pass 2 finishes, index construction is complete
and the index cache is written to disk.

Note the new revision of the help browser narrows indexing to just the
-help.pd patches. Previous versions would index all .pd files. This may be
subject to review, but the new indexing scheme is faster and produces less
noise in the earch results (i.e., you won't see any helper abstractions or
other patches which just happen to be bundled with the help patches).

Gabriela Bittencourt authored 3 years ago and Albert Gräf committed 3 years ago

This adds clickable links (employing calls to pdgui.doc_open) to the
help patches for related objects, if we can locate them. To support
this, we add an initial scan of the help patches, recording their names
and paths in the search index so that we can quickly look up their
locations.

NOTE: At present, the initial scan only covers the default hierarchy in
libdir, regardless of what the actual scope of the browser set in the
gui preferences and the help path is. Meta data will only be retrieved
for the help patches discovered in the initial scan.

Gabriela Bittencourt authored 3 years ago and Albert Gräf committed 3 years ago

If an object has a "related objects" field in its help patch meta data,
then the list of related objects will be added to the help index and
will be displayed along with its description in the help browser.

Moreover, objects will be listed in the search results if they have the
search term as a related object. For instance, a search for 'pulse' will
also list 'metro' because 'metro' has 'pulse' as a related object.

Note that depending on the actual meta data, relatedness isn't
necessarily symmetric. E.g., 'pulse' doesn't list 'metro' as a related
object in turn and thus a search for 'metro' will *not* show 'pulse' in
the search results, even though 'pulse' is related to 'metro'.

Gabriela Bittencourt authored 3 years ago and Albert Gräf committed 3 years ago

These interfere with incremental search (if an object name happens to be
a stop word then it won't show in the search results), so we get rid of
them.

pull from remote

Revert "added classname in m_obj.c"

See merge request ishankaler/purr-data!1

This reverts commit be6b2692