Chained has attribute treated as invalid syntax

[musjj]

The has attribute can be chained:

$ nix eval --expr "{} ? foo ? bar"
false

But nixfmt fails to format it:

$ nixfmt <<<"{} ? foo ? bar"
<stdin>:1:10:
  |
1 | {} ? foo ? bar
  |          ^
unexpected '?'
expecting '.', end of input, or operator

Meanwhile, single has attributes or dot operators work just fine:

$ nixfmt <<<"{} ? foo"
{ } ? foo

$ nixfmt <<<"{} ? foo.bar"
{ } ? foo.bar

[MattSturgeon]

I agree, this is a correctness issue, thanks for raising it.

However, I will also note that chaining ? after an ? expression will always evaluate to false, because the lefthand ? will evaluate to a boolean, and booleans don't have attributes.

Based on that, I don't see this as a high-priority issue, unless we discover some more useful examples that nixfmt fails to parse. PRs fixing correctness issues will always be welcome, regardless.

Testing locally, it seems like more useful expressions are handled correctly:

$ echo "{} ? foo ? bar" | nix run nixpkgs#nixfmt -- -
<stdin>:1:10:
  |
1 | {} ? foo ? bar
  |          ^
unexpected '?'
expecting '.', end of input, or operator

$ echo "{}.foo or { foo = "foo"; } ? foo" | nix run nixpkgs#nixfmt -- -
{ }.foo or { foo = foo; } ? foo

The error expecting '.', end of input, or operator is interesting; it hints that maybe ? tokens aren't considered to be "operators" internally. 🤔