Compose: allow overlapping sequences

[wismill]

Motivation

Currently xkbcommon does not support Compose overlapping sequences, unlike GTK and ibus. It means a sequence that is a prefix to a longer one is simply discarded with a warning.

This is unfortunate:

• There is discrepency for users of xkbcommon, e.g. Qt-based apps.

• It is impossible to have a sequence that is the prefix of another one. As a consequence, if one imports e.g. the system locale Compose file (as many – most ? – custom Compose files do), some sequences become impossible.

  Example: I used <Multi_key> <minus> <period> : "•" U2022 but upstream change in libX11 added <Multi_key> <minus> <period> <e> : "ė̄", so I had to introduce an alternative sequence for Qt apps, although the previous one works in GTK-based apps.

Proposed change

This commit introduces the new following API:

• Macro XKB_COMPOSE_HAS_OVERLAPPING_SEQUENCES_SUPPORT to test overlapping sequences support at compile time via #ifdef.
• xkb_compose_compile_flags: XKB_COMPOSE_COMPILE_OVERLAPPING_SEQUENCES: Allow overlapping sequences
• xkb_compose_status:
  • XKB_COMPOSE_CANDIDATE: A complete sequence has been matched, but a longer sequence also exists.
  • XKB_COMPOSE_CANDIDATE_ACCEPTED: The last sequence was accepted due to an unmatched keysym.

Fixes #395

NOTE: I intend to add the first two commits in #399. Will need to rebase once done.

Trying the patch

You can try it with

meson devenv -C build xkbcli-interactive-wayland --enable-compose --enable-compose-overlapping

or replace wayland with x11 if relevant.

Implementation in Qt

I asked feedback to Qt devs.

I sent also sent a merge request.

[wismill]

The idea is to create nodes that contain the overlapping results using a previously unused slot.

Possible enhancements:

• Naming
• Compose sequence iterator
• Doc
• Integration in toolkits (Qt in particular)

It would be great to be able to test this in Qt apps to test the design, but I will not do it myself.

[wismill]

I asked feedback to Qt devs.

[wismill]

Not too happy with the XKB_COMPOSE_CANDIDATE_ACCEPTED state. But if we return just XKB_COMPOSE_COMPOSED, then the client should handle the last keysym as not part of the sequence.

[wismill]

I sent a merge request in Qt.