Referring to a named watch, then adding, triggers sqlite crash

[ChrisPenner]

Referring to a named watch, then adding, triggers sqlite crash

> named = 18

x = named

.> add

  ⍟ I've added these definitions:

    x : Nat


  I added some default metadata.

Encountered exception:
SqliteQueryException
    { sql = "SELECT object.id
            FROM object
            JOIN hash ON object.primary_hash_id = hash.id
            WHERE hash.base32 = ? COLLATE NOCASE"
    , params = ""4cmnuksk66811n0hde9oroe54q1ku0r11fj50hvnvpfn7baooolpc4kpkocb3j6e6r7tu485mkvit586io4m4ua43ukd54i1kk24rpo""
    , exception = ExpectedExactlyOneRowException
        { rows = "[]" }
    , callStack = []
    , connection = Connection
        { name = "main"
        , file = "/Users/cpenner/.unison/v2/unison.sqlite3"
        }
    , threadId = ThreadId 295
    }

[aryairani]

This is still an issue as of today even with #4438 merged

[aryairani]

I'm guessing we just universally exclude watch expressions, but we should include them if a regular expression depends on them, or complain and fail when we try to add a dependent; but why not just include them?

Could just do a time-boxed initial investigation

[aryairani]

We hope that nobody is referring to named watches in their other definitions, but as an error message, an unformatted sqlite exception is p. bad. The problem is likely that we unconditionally exclude watch expressions when adding, even if they are dependencies.

Mitchell says maybe stop supporting named regular watches at all.

Chris wonders if it wouldn't be the same amount of work to fix the exclusion vs disallowing named regular watches.

test> named = []
x = named

followed by add produces the same-looking sort of crash