Signature Help should provide signature in the `do` notation

[fendor]

It looks like no signature help is provided when the AST node is in a do notation block:

You can see in the logs, the rules were triggered, but no signature help was provided.
If I change the code in question to not use do-notation, then signature help works as expected:

Happened on GHC 9.10.3

[fendor]

cc @jian-lin, this is probably just a matter of expanding the correct AST nodes, I assume?

[jian-lin]

Nice catch. I can have a look tomorrow the day after tomorrow.

this is probably just a matter of expanding the correct AST nodes, I assume?

I am not sure currently. Dumping the AST (haskell-language-server hiedb dump) may help.

[jian-lin]

Facts

I can reproduce this in GHC 9.10.2 and 9.12.2. This issue does not exist in GHC 9.6.7 and 9.8.4.

The direct cause is a HIE AST change introduced in 9.10. Here is the HIE AST node of expected (Node@/path/to/haskell-language-server/hls-test-utils/src/Test/Hls.hs:400:47-54:) in different versions.

GHC 9.6.7 and 9.8.4:

Source: From source
(annotations: {(HsVar,
                HsExpr)}), 
(types: [413]), 
(identifier info: {(name expected,  Details:  Just 413 {usage})})}

GHC 9.10.2 and 9.12.2:

Source: From source
{(annotations: {(HsVar,
                 HsExpr)}), 
 (types: []), 
 (identifier info: {(name expected,  Details:  Nothing {usage})})}
Source: generated by ghc
{(annotations: {(HsVar,
                 HsExpr)}), 
 (types: [437]), 
 (identifier info: {(name expected,  Details:  Just 437 {usage})})}

We can see that, besides the expected identifier From source, there is a new expected identifier generated by ghc. And all type info is moved to the generated by ghc identifier from the From source identifier.

Next step

Before modifying related code in HLS to fix this, I think we should first figure out if this HIE AST change is intended or is a GHC bug. Mayebe @wz1000 can share some pointers.

[vidit-od]

I tried working on this issue,

as recommended by @jian-lin , the type information for all of the Not working cases were stored in source: generated by ghc nodes. Enough time has also passed since the issue was raised so we can confirm that it is not a GHC bug. ( if it is fixed we can revert back to the old traversal maybe ? )

I have a working solution for this, hopefully it is ok if i raise a PR.

[jian-lin]

I tried working on this issue, as recommended by @jian-lin.

I do not think I have recommended you to working on this issue. That being said, PR from anyone is welcome.

This issue is on my radar. Personally, I would like to figure out which GHC commit causes this issue before fixing it in HLS. However, I have not got the time to bisect GHC yet.

[vidit-od]

I do not think I have recommended you to working on this issue.

sorry, i think i phrased it incorrectly.
By recommendation part, i mean all of your work on going through the AST Dump !
I have raised a PR for this, Feel Free to close it in case it is being explored by you.

Thank you !