internal: add failing incremental test

[matklad]

No description provided.

[matklad]

@jonas-schievink currently this returns 4 rather than 1. That is, we recalculate item_tree for file itself, as well as for each of the macro calls. Am I correct that we don't expect this happening?

[jonas-schievink]

yeah that shouldn't happen

[matklad]

Current guess: it's to_fragment_kind

[jonas-schievink]

ItemTrees for macro files seem to have a lot of unexpected query dependencies:

[2021-05-05T13:31:26Z DEBUG salsa::derived::slot] read_upgrade(FileItemTreeQuery(HirFileId(MacroFile(MacroFile { macro_call_id: LazyMacro(LazyMacroId(0)) })))): inputs=Tracked {
        inputs: [
            parse_macro_expansion(MacroFile { macro_call_id: LazyMacro(LazyMacroId(0)) }),
            intern_macro(MacroCallLoc { def: MacroDefId { krate: CrateId(0), kind: Declarative(InFile { file_id: HirFileId(FileId(FileId(0))), value: FileAstId::<syntax::ast::node_ext::Macro>(0) }), local_inner: false }, krate: CrateId(0), kind: FnLike { ast_id: InFile { file_id: HirFileId(FileId(FileId(2))), value: FileAstId::<syntax::ast::generated::nodes::MacroCall>(0) } } }),
            parse(FileId(2)),
            ast_id_map(HirFileId(FileId(FileId(2)))),
            parse(FileId(0)),
            ast_id_map(HirFileId(FileId(FileId(0)))),
            macro_def(MacroDefId { krate: CrateId(0), kind: Declarative(InFile { file_id: HirFileId(FileId(FileId(0))), value: FileAstId::<syntax::ast::node_ext::Macro>(0) }), local_inner: false }),
            macro_arg_text(LazyMacro(LazyMacroId(0))),
            hygiene_frame(HirFileId(FileId(FileId(0)))),
            hygiene_frame(HirFileId(FileId(FileId(2)))),
            ast_id_map(HirFileId(MacroFile(MacroFile { macro_call_id: LazyMacro(LazyMacroId(0)) }))),
        ],
    }

I would expect just 2: parse_macro_expansion and ast_id_map

[jonas-schievink]

Seems like these are all from Hygiene::new, which calls HygieneFrame::new

[jonas-schievink]

Making Hygiene::new go through another query does not seem to fix the bug (but costs us 100 MB of RAM)

[jonas-schievink]

Ah, that's probably just because HygieneInfo contains TextSizes, so its value always changes when some text is added or deleted in the file containing the macro invocation. Not really sure what the best way to fix this would be, maybe only storing AstIds in the hygiene data works? cc @edwin0cheng

[edwin0cheng]

Ah, that's probably just because HygieneInfo contains TextSizes, so its value always changes when some text is added or deleted in the file containing the macro invocation. Not really sure what the best way to fix this would be, maybe only storing AstIds in the hygiene data works? cc @edwin0cheng

In theory yes, but HygieneInfo mainly used inside lower_path and it is db independent such that we can't simply convent that back to offset of the HirFile.