Idempotency problem for unit of measure

[dsyme]

Issue created from fantomas-online

Formatted code

type herth = /second / second

Reformatted code

type herth = /second / second / second / second

Problem description

Fantomas was not able to produce the same code after reformatting the result.

Extra information

• The formatted result breaks my code.
• The formatted result gives compiler warnings.
• I or my company would be willing to help fix this.

Options

Fantomas master branch at 2022-05-18T11:51:41Z - ce92e66

Default Fantomas configuration

_{Did you know that you can ignore files when formatting from fantomas-tool or the FAKE targets by using a .fantomasignore file?}

[nojaf]

Hello, thank you for reporting this problem.
This is an interesting case actually.
Looking at the AST:

TypeAbbrev
 (Ok,
  Tuple
    (false,
     [(true,
       StaticConstant (Int32 1, tmp.fsx (2,13--2,21)));
      (false,
       LongIdent (SynLongIdent ([second], [], [None])))],

Notice that StaticConstant (Int32 1, tmp.fsx (2,13--2,21)), that isn't the best representation as there is no 1 written anywhere.

The range of the SynConst.Int32 does include that entire abbreviation /second and due to some recent change, we print numbers straight from the ISourceText.

fantomas/src/Fantomas.Core/CodePrinter.fs

Lines 4328 to 4334 in ce92e66

	and loopTTupleList =
	function
	| [] -> sepNone
	| [ (_, t) ] -> loop t
	| (isDivide, t) :: ts ->
	loop t -- (if isDivide then " / " else " * ")
	+> loopTTupleList ts

loop t will eventually print the SynConst and repeat the / second.

It could be changed to something like:

    and loopTTupleList =
        function
        | [] -> sepNone
        | [ (_, t) ] -> loop t
        | (isDivide, t) :: ts ->
            match t with
            | TStaticConstant(SynConst.Int32 1, _) when isDivide -> !- "/ " +> loopTTupleList ts
            | _ -> loop t -- (if isDivide then " / " else " * ") +> loopTTupleList ts

However, maybe to better capture this, we could introduce a special case in SynType.

That workaround will probably do the trick.
TypeDeclarationTests.fs would be a good place to add the unit test.

Are you interested in submitting a PR?

[dsyme]

Looks like we should fix the SyntaxTree here

[dsyme]

Are you interested in submitting a PR?

At some point - it will be recorded and linked as a limitation in .fantomasignore of dotnet/fsharp, and we will eventually clear those out.

[dsyme]

See also #2265, whenever we fix these we should likely do them at the same time :)