Regression in autocomplete for Julia 1.10

[nilshg]

I feel like I discussed this with people before but there isn't an issue:

~> julia +1.9 -q 

julia> nt = (col1 = 5, secondcol = 6)
(col1 = 5, secondcol = 6)

julia> nt.col1 = nt.sec<TAB> # autocompletes

julia> exit()

~> julia +1.10 -q

julia> nt = (col1 = 5, secondcol = 6)
(col1 = 5, secondcol = 6)

julia> nt.col1 = nt.sec<TAB> # crickets

Still broken on 1.11.3, so I assume this is a JuliaSyntax thing?

[fingolfin]

I can confirm this is also broken on master. Note that the REPL completion code also underwent a major refactoring, and IIRC (but I might not) this was between 1.9 and 1.10. So that could also be responsible. Perhaps @aviatesk or @IanButterworth know better?

[IanButterworth]

There are unfortunately a variety of completions regressions: https://github.com/JuliaLang/julia/issues?q=sort%3Aupdated-desc%20is%3Aissue%20is%3Aopen%20label%3Acompletions%20label%3Abug

I've looked into trying to fix them, but not been able to make progress. The regressions seem mostly caused by the improvements @aviatesk made because we didn't/don't have as much test coverage as we should have for completion behavior.

[aviatesk]

Agh, this isn't good. I didn't expect so many regressions. I think we have a few options:

1. Revert the original commit: But my changes were made over a year (or years?) ago, so reverting now might be tough.
2. Patch each issue individually: It's tedious, but that's probably this is exactly how we've improved REPL completion quality so far.
3. Use something like JuliaSyntax to more intelligently parse and detect the code we're trying to complete.

Option 1 probably isn't realistic, so we're likely stuck with 2 or 3.
Personally, I'd prefer to go with option 3. I'm actually working on a new completion engine for another project, so in the near future, I might be able to apply what I learn there to option 3.
But I'm currently swamped with that project and don't have the bandwidth to dedicate time to individual issues in JuliaLang/julia right now.
So, if anyone else has time to work on option 2 or 3, that would be a huge help.

[nilshg]

I'll just add that someone on Slack had a theory that autocomplete is now "smart" and won't autocomplete the nt.sec in my example because NamedTuples are immutable and so I can't assign to nt.col1 anyway. They correctly pointed out that

nt.col1 == nt.sec

does autocomplete. I'm not sure whether that's true, but I'll note that

julia> nt = (; col1 = rand(2), secondcol = rand(2));

julia> nt.col1 .= nt.sec

also doesn't autocomplete, despite nt.col1 being mutable.

Finally, my issue was mainly motivated by DataFrames type workflows like

julia> df = DataFrame(col1 = rand(2), secondcol = rand(2));

julia> df.col1 = df.seco

which are also broken.