Binds of a HsLet ignore the location of HsLocalBinds

[ocharles]

To recap our conversation on IRC:

11:47 <ocharles> alanz, mpickering: I'm looking at doing pretty printing with just ghc-exactprint - essentially just moving DPs around to get the formatting I want. It's pretty cool! But I have a q... ghc-exactprint seems to discard some locations, and I wonder why
11:47 <ocharles> for example, Annotate for HsLet in HsExpr ignores the location of the binds of the let statement, so this doesn't do anything:
11:48 <alanz> ocharles: have you looked at brittany? Does a similar thing
11:48 <ocharles> case expr of HsLet binds body -> setEntryDPT binds (DP (100000, 0))
11:48 <ocharles> alanz: yep, I know of brittany, but brittany actually does the printing itself
11:48 <alanz> ocharles: which GHC version?
11:48 <ocharles> I wanted to do nothing except move deltas around
11:49 <ocharles> 8.0.2
11:50 <ocharles> alanz: but https://github.com/alanz/ghc-exactprint/blob/master/src-ghc82/Language/Haskell/GHC/ExactPrint/Annotater.hs#L1877 seems problematic, it just discards the srcspan of the binds
11:51 <alanz> ok, you need to look in the src-ghc80 dir
11:51 <alanz> which is the same
11:51 <alanz> also discards the location
11:51 <ocharles> right
11:52 <alanz> That location was added to allow indexing the list by it, so it could move independently
11:52 <alanz> but before that the engine was set up to ignore it
11:52 <alanz> and I have not gone back to make use of it yey
11:52 <alanz> yet
11:52 <ocharles> ok, so this could be considered a bug?
11:52 <ocharles> hsmodName is similarly affected, setting the entry DP of that doesn't do anything
11:52 <alanz> There was some technical issue about managing the list in the surrounding scope, and breaking existing behaviuour
11:53 <ocharles> and I haven't tried transforming much more than that :)
11:53 <alanz> Well, it is more of a feature at the moment :(
11:53 <alanz> The thing it to look at the Annotater code and see what is used, and work from there
11:54 <alanz> You can also call showAnnData to see exactly what annotations exist
11:54 <ocharles> my use case is I want to "swing" bindings of a let statement under the let, so `let x =` becomes `let\n<space><space>x =`. Right now I'm doing that by using the first location of the binds, but that breaks if the let statement opens with a type signature
11:54 <alanz> What I normally do is run a test with the "from" source code, look at what I get, then with the manually formatted new one, and compare
11:55 <alanz> And that should be possible (it is a design goal), but you need to check what to work with
11:55 <alanz> (my brain is a bit rusty)
11:56 <alanz> iirc, I originally wanted the indent to be on the location of the binds, then each could start at zero and be freely re-arranged
11:56 <alanz> But that is not what ended up in the code, and it needs to be fixed
11:57 <alanz> The "setLayoutFlag" triggers the indentation, by the way
11:57 <alanz> no, sorry, markLocalBindsWithLayout
11:58 <ocharles> so far I haven't written my own interpreter for Annotate because I'm hoping the Print module should be sufficient
11:58 <alanz> it should be
11:59 <alanz> The problem you are seeing comes because the HsLet binds are not decls, so are split into sigs and binds
11:59 <alanz> which means you end up with this: https://github.com/alanz/ghc-exactprint/blob/master/src-ghc80/Language/Haskell/GHC/ExactPrint/Annotater.hs#L1738
11:59 <ocharles> well, that would be one explanation. my expectation was that each dp of the binds (and sigs) are relative to the dp of the whole HsLocalBinds
12:00 <alanz> that is what I want to see too.
12:00 <alanz> Unfortunately it did not happen originally, and needs to be retrofitted to be that way
12:00 <ocharles> ok, if we're in agreement there then I will at least open an issue
12:00 <alanz> please do