[WIP] Refactor auto completion

[stefan-kolb]

Major changes in this PR:

• Only use auto-completion when at the end of a word (space after)
• Detection modes: Only use case-sensitive detection for now (performance, logic, we can re-add functionality from here)
• Remove method getPrefix() - Intention of this method was unknown
• Use Nameautocompleter for every PERSONNAME BibTeX field, not only author & editor

[stefan-kolb]

The current auto completion has two modes:

• a case sensitive mode
• a case insensitive mode

The case sensitive mode is activated when one types a word that has mixed case, e.g., gZhhUh.
If the word is only lowercase the auto completion will perform a case insensitive mode.

The case sensitive mode is straight-forward: It will lookup a word that can be completed from the current partial word.
Case insensitive lookup tries to find a word that can be completed from the input based on ignore case search. If it found one, it will still present the word in its original casing, e.g., some -> Something

Discussion point
What is the purpose of the case-insensitive mode? The detection (only lowercase in word) seems arbitrary to me. How should the completion be performed in general or what are typical algorithms for this? I would either suggest case-sensitive or a possibility to configure this but no auto detection?!

All Shells (UNIX) are case sensitive by default.

[stefan-kolb]

Ok, I got something working here now. Massively refactored the code (god was this hard to read 😭).
I'm still not quite happy with the result, but I hope this will be a starting point for more improvements with a readable codebase.

@JabRef/developers WDYT? I'd be happy if you try the version with a few use cases of your own to check whether it is working correctly for you. http://builds.jabref.org/auto-complete/

[simonharrer]

make suggestions final and unmodifyable.

[simonharrer]

LGTM 👍

[stefan-kolb]

@JabRef/developers Does it work for you? Should we merge this in or any further suggestions?

[oscargus]

Would it make sense for the NameAutoCompleter to work for all fields with BibtexSingleFieldProperties.PERSON_NAMES set? I do not really know how it works today, but I seem to recall that it wasn't obvious if the same AutoCompleter was used for both "author" and "editor". (Haven't had time to test it yet.)

[stefan-kolb]

Would be possible, currently it only works on author and editor.

if ("author".equals(fieldName) || "editor".equals(fieldName)) {
             return new NameFieldAutoCompleter(fieldName, preferences);

[stefan-kolb]

I quick checked. Currently only those two fields have the property BibtexSingleFieldProperties.PERSON_NAMES set.

[oscargus]

Correct but there's is a PR (don't know the number now) which adds 20 or so fields.

[stefan-kolb]

Ah ok. We could do something like this:

if (InternalBibtexFields.getFieldExtras(fieldName).contains(BibtexSingleFieldProperties.PERSON_NAMES)) {
            return new NameFieldAutoCompleter(fieldName, preferences);
        }

But I get NullpointerExceptions in the tests due to the InternalBibtexFields relying on Globals.prefs (InternalBibtexFields.java:125)

[oscargus]

OK! These dependencies are annoying... Mock JabRefPreferences?

[koppor]

What is the purpose of the case-insensitive mode? The detection (only lowercase in word) seems arbitrary to me.

Is this still a valid question? If yes: I was thinking that casing causes more keys to type (shift key). Thus: being lazy should instruct JabRef to search for everything. Being non-lazy (capital letter), I want the tool to search only for these cases as I demand something specific.

In powerline-shell or Windows CMD, the search IMHO also is case-insensitive.

I have not researched for typical algorithms the recent time. When refactoring the auto completion a few years ago, I found nothing useful. However, I did not document my search.

[koppor]

Quick test:

• author field
• Kwantes, PieterM. and Van Gorp, Pieter and Kleijn, Jetty and Rensink, Arend
• Change V in Van Gorp to v
• go to the end of the line, type and v.
• One cannot type v
• title field
• Text: Towards Compliance Verification Between Global and Local Process Models
• Go to the end of the entry
• type and
• about is displayed and the accepted characters are abo and not the typed and.

The typed characters have to be kept! The autocompletion must not override the user's input.

I can send you my large bib file privatly. Don't want to share it publicly.

[koppor]

Exception in thread "AWT-EventQueue-0" java.lang.StringIndexOutOfBoundsException: String index out of range: -67
        at java.lang.String.substring(String.java:1931)
        at net.sf.jabref.gui.autocompleter.AutoCompleteListener.insertKeyAndSuggestion(AutoCompleteListener.java:148)
        at net.sf.jabref.gui.autocompleter.AutoCompleteListener.keyTyped(AutoCompleteListener.java:104)
        at java.awt.Component.processKeyEvent(Component.java:6490)

[koppor]

This PR is superseeded by #1136