[platform-text-dev] Fw: info for implementors of ICompletionProposal

["Claude Knaus" <Claude_Knaus@xxxxxxx>]

----- Forwarded by Claude
Knaus/ZRH/OTI on 25.11.2002 10:13 -----

Claude Knaus 22.11.2002 19:16

To:        platform-text-dev@xxxxxxxxxxx         cc:                 Subject:        Fw: info for implementors of ICompletionProposal

----- Forwarded by Claude
Knaus/ZRH/OTI on 22.11.2002 19:22 -----

Claude Knaus 22.11.2002 18:25

To:        platform-text-dev@xxxxxxxxxxx         cc:                 Subject:        info for implementors of ICompletionProposal

A heads up for implementors of ICompletionProposal
(used by content assist):

When a completion proposal applies a
change to a document, it usually is the case
that only these document changes are
actually applied to the document.
However, in certain cases (like when
being in linked position mode - used in templates and
others) a document change can trigger
other document changes. Up to now, this
was not of concern, since these document
changes always happened to the
right of the original document change.

Due to a new change in the linked position
mode, it is now possible to trigger document
changes which are to the left of the
originating document change. The effect is that
if the resulting selection calculated
based on the assumption that no additional
document changes are triggered to left
of the original document change, the
selection might become invalid.
In concrete, this can happen if a template
is active and content assist is issued
on a position, which is not the left
most of its position type.

To correctly implement ICompletionProposal.getSelection(),
a document listener should be
installed during ICompletionProposal.apply()
to keep track of possibly additional affecting
document changes. This can also be done
by inserting a document position with an
apropriate document position updater
(see JavaCompletionProposal for how this can
be done). The idea is to track a hopefully
'stable' reference point during the document
changes. The selection should then be
calculated with respect to this reference point.

Note: This is not considered as an API
change. It's just that so far, the API was underspecified.

-- Claude