Ragnar, I do not believe we wish to use keywords interchangeably with symbols for binding keys.

Furthermore, I think this is a good patch that solves the problems introduced in CLJ-1318 (I'll take the blame for those!).

I thought so, but in writing this reply I realised a smaller change
might be possible, depending on the desired behaviour/outcome.

CLJ-1318 ("Support destructuring maps with namespaced keywords")
introduced two potentially undesired behaviours:

1. Namespaces were removed from symbols occurring in any binding key
position regardless of whether they were used in map destructuring
or not. This lead to the behaviour initially reported in this
issue; the illusion of being able to bind a qualified name in a
non-map-destructuring context:

This currently "works" because namespaces are unconditionally removed
from symbols. However:

throws because destructuring logic is bypassed if every binding key is
a symbol.

2. Keywords were allowed in any binding key positions. Keywords are
converted to symbols (retaining namespace) and treated according to
the rules of symbols in the rest of the destructuring logic. From
what I understand the idea was to allow keywords only in map
destructuring, but again the change change was effected for any
binding key. This leads the following:

A top-level check was put in place to (from my understanding) try and
prevent the usage of keywords in such positions:

However, this check only checks the top level binding entries, and
fails the recursive nature of destructuring (as demonstrated by
example above).

The current patch makes the following changes:

1. Revert problematic changes introduced by CLJ-1318:
1.1 revert unconditional removal of namespace from any namespaced symbol encountered
1.2 revert acceptence of keywords in any binding key position
1.3 revert insufficient check for keywords used in "illegal" positions.
2. Re-implement support for namespaced symbols, and keywords, but only
in the context of map destructuring.

If that is what we want to accomplish I believe the patch is the
smallest possible. However, if the keyword behaviour is actually
desired (basically allowing keywords being used interchangably with
symbols for binding keys) then the patch could be smaller.

Please excuse the rather lengthy reply.

Is this the smallest change that can be made?

Updated patch with two changes:

1. Removed initial check for keywords in binding keys as that only looked for keywords at top-level and failed to catch cases like:

Keywords in non-map destructuring binding positions (like the example) will now fail with "Unsupported binding form: :x" instead. This is a change from the current behaviour where only top-level keywords would be caught, but the current exception is the slightly more specific "Unsupported binding key: :x". If a better, more specific exception is required, I'd be happy to update the patch.

2. Tests: added test for regression, updated/amended existing tests.

I can confirm that this is a regression from CLJ-1318. After that change, namespaces were removed from symbols regardless of whether they were used as part of a map destruring or not. The only reason the exception is caught in the first test case is because that binding form has no destructuring at all, in which case the destructuring logic is bypassed.

I've attached a patch that moves the namespace removal (and keyword handling) into `pmap` instead as it's really a special case for map destructuring only.