v0.9 no longer displays the name of the duplicate key and table on parse

[zanieb]

See https://github.com/zanieb/toml-regression-1 for a simple reproduction

In 0.8.19

TOML parse error at line 4, column 1
  |
4 | python-multipart = "https://example.com/bar.whl"
  | ^
duplicate key `python-multipart` in table `tool.uv.sources`

In 0.9.2

TOML parse error at line 4, column 1
  |
4 | python-multipart = "https://example.com/bar.whl"
  | ^^^^^^^^^^^^^^^^
duplicate key

It seems okay to avoid repeating python-multipart but the "table ..." part seems pretty helpful.

Originally discovered at https://github.com/astral-sh/uv/pull/14571#discussion_r2213787884

[epage]

The duplicate key was dropped because at that time, the parser didn't support allocations for errors and I felt that being able to show the context would make up for it enough that this didn't motivate adding allocations.

Showing which table it was in felt redundant (since we show the location). If the case can be made for it, it will require expanding our error reporting scheme as it doesn't support that today. What it does support is the ability to highlight other sections in the file. We aren't currently exposing this or using it in our own rendering. In some cases, what the parser is providing the span for is the location of the first key that the current key is a duplicate of.

[zanieb]

Makes sense, I figured it was motivated by changes like that :)

Showing which table it was in felt redundant (since we show the location)

While it's technically redundant, I think it's less obvious where to go from just the line numbers. I think it's hard to argue it's not more user friendly to display the table name in some way, though I think it's fair to argue it's not worth the effort.

What it does support is the ability to highlight other sections in the file.

That could definitely be a nice approach too, though I think it can be hard to get right.

No pressure to make any changes here, just wanted to make sure you were aware these changed.