infix_spaces_linter lacks a way to specify different rules for different `=` (`EQ_ASSIGN`, `EQ_SUB`, and `EQ_FORMALS`)

[MichaelChirico]

Some authors (e.g. data.table) apply different rules for different =, for example:

f = function(a=1) a
f(a=2)

Specifically, EQ_SUB and EQ_FORMALS should not have surrounding spaces, but EQ_ASSIGN should.

Currently, exclude_operators only supports string inputs, so this type of node-level differentiation is impossible.

There are two options:

(1) Allow specifying node names in exclude_operators=. exclude_operators = c("EQ_SUB", "EQ_FORMALS") is unambiguously a parse_tag. The advantage is a simplified user interface, the disadvantage is overloading.
(2) Add a new argument exclude_parse_tags=, which will match against lintr:::infix_metadata$parse_tag instead of lintr:::infix_metadata$string_value.

[AshesITR]

I'm slightly leaning towards (1) so as to not add two arguments that can do "the same". Are we exposing the xml tag names somewhere, so users can learn about them without studying xmlparsedata?

[MichaelChirico]

That feels like exposing some implementation details... e.g. we don't expose lintr:::style_regexes for object_name_linter() (or should we?).

I'd suggest pointing to utils::getParseData() and accepting the parser node names instead. Otherwise we could point to xmlparsedata:::xml_parse_token_map as a more canonical source.

[AshesITR]

Okay; "EQ_SUB", "EQ_FORMALS" and "EQ_ASSIGN" are the names for both, getParseData() and xmlparsedata().
Pointing to getParseData() will only differ in cases without ambiguity.