hsdev

Haskell development library and tool with support of autocompletion, symbol info, go to declaration, find references, hayoo search etc. Uses fsnotify to watch for changes.

Installation

Stack

To install latest version with LTS resolver (14.10 should be fine), add these extra packages to your global stack project:

resolver: lts-14.10
extra-deps:
- hsdev-0.3.3.5
- haddock-api-2.21.0  # not required if you disable `docs` flag
- hdocs-0.5.3.1  # not required if you disable `docs` flag

And install with stack install hsdev

If you want to use latest nightly resolver and ghc-8.8, then there are some more extra deps required:

resolver: nightly-2019-10-19
extra-deps:
- hsdev-0.3.3.5
- Cabal-3.0.0.0
- haddock-api-2.23.0  # for `docs` flag
- haddock-library-1.8.0  # for `docs` flag
- hdocs-0.5.4.0  # for `docs` flag
- direct-sqlite-2.3.24
- git: https://github.com/mvoidex/sqlite-simple  # actual version is broken
  commit: cc94f6e303b19aeaed2ac21cbccf6f5c7b74274d
- semigroups-0.18.5

Install with stack install hsdev

Alternatively, you can git clone https://github.com/mvoidex/hsdev and install with stack install or stack install --stack-yaml nightly.yaml

Cabal

$ cabal sandbox init
$ cabal new-install hsdev

Usage

Main idea is to hold in memory scanned sourced and installed modules, so that getting info about symbols and modules is fast. It also doesn't require much work to integrate with some editor:

1. Create hsdev run ... process
2. Connect to it via socket
3. Send scan command with paths/files once
4. Use other commands to get completions, modules and symbols info; check and lint sources

Editors

1. SublimeText: SublimeHaskell plugin
2. Atom: work in progress, atom-haskell-hsdev plugin

Start server

Use hsdev start to start remote server.
If you installed hsdev with stack, run server with stack exec -- hsdev start command, so that correct $GHC_PACKAGE_PATH is set where hsdev library installed.
Client commands then doesn't require running them with stack exec.
Specify --db file.db, where hsdev will store information (SQLite database, see hsdev.sql for schema).
Then you can connect to server and send requests (see requests/responses) or you can use hsdev itself. It will send command to server and outputs the response. Scan sources, installed modules and you're ready to request various information: scope, completions, info about symbols etc.

Typical usage is:

PS> stack exec -- hsdev start
Server started at port 4567
PS> hsdev scan project ~/projects/hsdev --stack --silent
[]
PS> hsdev complete runC -f ./src/HsDev/Server/Commands.hs | jq -r '.[] | .id.name + \" :: \" + .info.type'
runClientM :: ServerM (ReaderT CommandOptions m) a
runConcurrently :: Concurrently a -> IO a

Stack support

hsdev uses stack to build dependencies and to get corresponding package-dbs. As long as hsdev uses ghc as library, it passes --compiler and --arch options (since 0.1.7.2) to stack in order to get compatible package-dbs built with same compiler.

Building without haddock/hdocs dependency

Disable flag docs to build without these dependencies: cabal configure -f-docs or stack install --flag hsdev:-docs

Commands

Run hsdev -? to get list of all commands or hsdev <command> -? (hsdev help <command>) to get detailed info.

• version — returns version number
• start, run and stop — server commands, start remote server, run server or stop remote server.
• connect — connect to server to send commands from command line (for debug)
• ping — ping server
• listen — connect to server and listen for its log (for debug)
• set-log — set log level
• scan — scan installed modules, cabal projects and files
• scan project, scan file, scan package-dbs — same as above, but with additional options
• set-file-contents — set data to use as file contents, used to work with unsaved files
• docs, infer — scan docs or infer types for sources
• remove, remove-all — unload data
• packages, projects, sandboxes — list information about specified modules, packages, projects or sandboxes
• symbol, module, project, sandbox — get info about symbol, module, project or sandbox
• whoat, whois, lookup — find project-visible or imported symbol
• scope, scope modules — get modules or declarations, accessible from file
• usages — find usages of symbol
• complete — get completions for file and input
• hayoo — search in hayoo
• cabal list — search packages info
• lint, check, lint-check — lint or check source files. These commands have some advantages over ghc-mod ones: lint uses hlint as library, check returns more precise source position and also uses project description to pass -package flags, these commands also can accept file contents
• types — get types for all source spans
• flags, langs — list ghc flags and language extensions
• ghc eval — evaluate expression
• ghc type — get type of expression
• autofixes — get suggestions to fix some warnings and errors
• rename — get regions to rename some symbol
• refactor — apply suggestions/renames from previous commands

TODO: Detailed commands description with examples

Scan

You can scan package-dbs, projects and files.

To scan global and user package-dbs, run

PS> hsdev scan package-dbs --global-db --user-db --silent
[]
PS> hsdev module -h -m System.IO | jq -r '.[].location.package'
base-4.12.0.0

To scan project use scan project, which will also scan dependens package-dbs (unless --no-deps flag set)

PS> hsdev scan project ~/projects/hsdev --stack --silent
[]
PS> hsdev module -f ~/projects/hsdev/src/HsDev/Client/Commands.hs | jq -r '.[].exports[] | .id.name + \" :: \" + .info.type'
runClient :: (ToJSON a, ServerMonadBase m) => CommandOptions -> ClientM m a -> ServerM m Result
runCommand :: ServerMonadBase m => Command -> ClientM m Value

To scan file use scan file:

PS> hsdev scan file ~/projects/hs/Main.hs --stack --silent
[]
PS> hsdev module -f ~/projects/hs/Main.hs | jq -r '.[].exports[] | .id.name + \" :: \" + .info.type'
main :: IO ()

Whois/whoat

Get information for symbol in context of source file. Understand qualified names and also names qualified with module shortcut (import ... as), note M. qualified for map, and local definition toResult:

PS> dev whois M.lookup --file .\src\HsDev\Client\Commands.hs | json | % { $_.id.name + ' :: ' + $_.info.type }
lookup :: Ord k => k -> Map k a -> Maybe a
PS> hsdev whoat 64 1 -f .\src\HsDev\Client\Commands.hs | json | % { $_.id.name + ' :: ' + $_.info.type }
toValue :: m a -> m Value
PS> hsdev whoat 55 32 -f .\src\HsDev\Client\Commands.hs | json | % { $_.id.name + ' :: ' + $_.info.type }
toResult :: ReaderT CommandOptions m a -> m Result

Usages

Returns all places where symbol is used

PS> hsdev usages Data.Map.toList | json | % { $_.in.name, $_.at.line, $_.at.column -join ':' }
Data.Deps:33:90
Data.Deps:57:62
HsDev.Symbols.Types:93:12
HsDev.Symbols.Types:95:54
HsDev.Symbols.Types:104:74
HsDev.Symbols.Types:104:94
...

AutoFixes/Rename/Refactor

Autofix commands used to assist for automatic fix of some warnings and hints from hlint. autofixes command parses check and lint command output, and returns corrections — data with regions and suggestions to fix. refactor command perform fix of selected corrections and also updates positions of other corrections, such that they stay relevant. These updated corrections can be used to pass them to refactor again. Example of interactive command, based on this command in SublimeHaskell:

Rename generates corrections to rename symbol

# Get corrections
PS> $corrs = hsdev check-lint ..\haskell\Test.hs | hsdev autofixes --stdin
# Apply first correction, other corrections passed as --rest param to update their positions
# Result is updated --rest corrections, which can be used again
PS> $corrs2 = ($corrs | jq '[.[1]]' -c -r) | hsdev refactor --pure --rest (escape ($corrs | jq '.[1:5]' -c -r)) --stdin
# One more
PS> $corrs2 | hsdev refactor --stdin --pure

Examples

PS> hsdev start --db hsdev.db
Server started at port 4567
PS> hsdev scan --cabal
{}
PS> hsdev scan --project hsdev
{}
PS> hsdev module --project hsdev -h | json | % { $_.name } | select -first 3
HsDev.Database.Update
HsDev.Client.Commands
HsDev.Scan
PS> hsdev symbol enumProject --src | json | % { $_.id.name + ' :: ' + $_.info.type }
enumProject :: Project -> m ScanContents
PS> hsdev complete tr -f .\tools\hsdev.hs | json | % { $_.id.name }
traverse
traverseDirectory
...
PS> hsdev stop
{}

Hayoo in GHCi

You can use this ghci.conf to allow search from ghci:

:set prompt "λ> "

import Control.Monad.Error
import HsDev.Tools.Hayoo
import Data.List (intercalate)

:{
let
    showHayooFunction f =
        (hayooName f ++ " :: " ++ hayooSignature f) :
        (map ('\t':) $
            lines (untagDescription (hayooDescription f)) ++
            ["-- Defined in '" ++ intercalate ", " (hayooModules f) ++ "', " ++ hayooPackage f])
    showHayoo = concatMap showHayooFunction . resultResult
:}

:def hayoo \s -> return $ "runErrorT (hayoo \"" ++ s ++ "\" Nothing) >>= (mapM_ putStrLn) . either (return . (\"Error: \" ++)) showHayoo"

Usage:

λ> :hayoo (a -> c) -> (b -> c)
query :: (a -> c) -> b -> c
        query f x walks the structure x (bottom up) and applies f
         to every a, appending the results.
        -- Defined in 'Text.Pandoc.Walk', pandoc-types
...

JSON

PowerShell

I'm using PowerShell and json function from PSUtils to parse JSON output, which returns PSObject, that can be inspected in common way:

PS> hsinspect module GHC -g "-package ghc" | json | % { $_.module.declarations } | % { $_.name + ' :: ' + $_.decl.type } | select -first 5
ABE :: id -> id -> HsWrapper -> TcSpecPrags -> ABExport id
ABExport :: data ABExport id
ACoAxiom :: CoAxiom Branched -> TyThing
AConLike :: ConLike -> TyThing
ATyCon :: TyCon -> TyThing

jq

Another way is to use jq tool, which is pretty simple:

PS> hsinspect module GHC -g "-package ghc" | jq -r '.module.declarations[range(0;5)] | .name + \" :: \" + .decl.type'
ABE :: id -> id -> HsWrapper -> TcSpecPrags -> ABExport id
ABExport :: data ABExport id
ACoAxiom :: CoAxiom Branched -> TyThing
AConLike :: ConLike -> TyThing
ATyCon :: TyCon -> TyThing