@hackage hie-bios0.6.3

Set up a GHC API session

hie-bios

hie-bios is the way to specify how haskell-language-server and ghcide set up a GHC API session.

Given a Haskell project that is managed by Stack, Cabal, or other package tools, haskell-language-server needs to know the full set of flags to pass to GHC in order to build the project. These flags might contain some compilation options like -O2, but a lot of the time they are package dependencies such as -package-id directory-1.3.6.0, which also need to be built beforehand. hie-bios satisfies both these needs.

Its design is motivated by the guiding principle:

It is the responsibility of the build tool to describe the environment which a package should be built in.

Using this principle, it is possible to easily support a wide range of tools including cabal-install, stack, rules_haskell, hadrian and obelisk without major contortions. hie-bios does not depend on the Cabal library nor does it read any complicated build products and so on.

How does a tool specify a session? A session is fully specified by a set of standard GHC flags. Most tools already produce this information if they support a repl command. Launching a repl is achieved by calling ghci with the right flags to specify the package database. hie-bios needs a way to get these flags and then it can set up a GHC API session correctly.

Further it means that any failure to set up the API session is the responsibility of the build tool. It is up to them to provide the correct information if they want the tool to work correctly.

Explicit Configuration

The user can place a hie.yaml file in the root of the workspace which describes how to set up the environment. There are several supported ways to setup the environment.

Stack

To explicitly state that you want to use stack, the basic configuration hie.yaml would look like:

cradle:
  stack:

This configuration suffices if your whole project can be loaded by the command stack repl. As a rule of thumb, this works if the project consists of only one executable, one library and one test-suite.

If your project is more complicated, you need to specify which components you want to load. A component is, roughly speaking, a library/executable/test-suite or benchmark in stack. You can view the components/targets of a stack project by executing the command:

$ stack ide targets

Since we have two test-suites, one executable and a library, for hie-bios, this would output the following:

$ stack ide targets
hie-bios:lib
hie-bios:exe:hie-bios
hie-bios:test:bios-tests
hie-bios:test:parser-tests

For an explanation of the target syntax, we refer to the documentation of the target syntax.

To tell hie-bios which component it should load, the following hie.yaml can be used:

cradle:
  stack:
    component: "<component name>"

where <component name> is the name of the component/target you want to load. While the component is optional, this is recommended to make sure the correct component is loaded.

Why is this not enough? Usually, you have multiple components with different dependencies. Your library won't depend on tasty or hspec, but your test-suite might. With this setup, you would only be able to load files from the given component.

Since you rarely only want to load a single component in a stack project, we have special syntax to be able to conveniently specify which directory belongs to which component. It is basically a multi-cradle.

cradle:
  stack:
    - path: "./src"
      component: "hie-bios:lib"
    - path: "./exe"
      component: "hie-bios:exe:hie-bios"
    - path: "./tests/BiosTests.hs"
      component: "hie-bios:test:hie-bios"
    - path: "./tests/ParserTests.hs"
      component: "hie-bios:test:parser-tests"

Here you can see two important features:

  • We provide a mapping from a filepath to component.
    • That way, we specify that a file such as ./src/HIE/Bios.hs belongs to the component hie-bios:lib.
  • The filepath can be a file.
    • This is convenient if components are overlapping.

This way we specified which component needs to be compiled given a source file for our whole project.

Debugging a stack cradle

If you find that hie-bios can't load a certain component or file, run stack repl and stack repl <component name> to see if stack succeeds in building your project. Chances are that there is a problem in your project and if you fix that, hie-bios will succeed to load it.

Also, see notes for testing your configuration.

Otherwise, please open an issue.

Ignoring directories

You can combine the multi-cradle with a none-cradle to ignore all source files in a certain directory. The syntax is a bit verbose:

cradle:
  multi:
    - path: "./tests/projects"
      config:
        cradle:
          none:
    - path: "./"
      config:
        cradle:
          stack:
            - path: "./src"
              component: "hie-bios:lib"
            - path: "./exe"
              component: "hie-bios:exe:hie-bios"
            - path: "./tests/BiosTests.hs"
              component: "hie-bios:test:hie-bios"
            - path: "./tests/ParserTests.hs"
              component: "hie-bios:test:parser-tests"

This way, we specify that we do not want to load any files in our test project directories.

Internal Libraries

Internal libraries are not well supported in stack. Since the syntax stack repl <internal library name> doesn't work, hie-bios will generally not work with internal libraries using stack.

Cabal

To use cabal, the basic explicit configuration looks similar to stack's configuration.

cradle:
  cabal:

The implication of this configuration is a bit different, though. Given a source file to load, we will use cabal repl <filename> to find the component of the given filepath.

This configuration should work in (almost) every standard project setup, since cabal finds the component associated to a given source file. However, due to an unfortunate bug, this fails on some files with cabal versions older than 3.4. So, to make your project loadable by older cabal versions, you can specify a component to load. A component is roughly speaking a library, executable, test-suite or benchmark in cabal. The hie.yaml file looks like this:

cradle:
  cabal:
    component: <component name>

This tells hie-bios that whichever source file it tries to load, this source file should be handled as if it belongs to <component name>.

As an example, to load the library of hie-bios, the following hie.yaml can be used:

cradle:
  cabal:
    component: "lib:hie-bios"

The component syntax "lib:hie-bios" refers to the library of the package hie-bios. For a complete reference of the component syntax, we refer to the documentation.

Note that cabal and stack have different ways of specifying their components.

If we only specify a single component, then we can only load source files from this component. This is unsatisfactory as we want to be able to navigate our project freely and work on multiple components (test-suite, library, executable, etc...) in parallel.

In a project such as hie-bios, we have more than one component, in particular we have four:

  • An executable
  • A library
  • Two test-suites

The component syntax can easily be extracted from the hie-bios.cabal file. Relevant sections are:

...
Name:                   hie-bios
...

Library
  ...
  HS-Source-Dirs:       src

Executable hie-bios
  ...
  Main-Is:              Main.hs
  HS-Source-Dirs:       exe

test-suite parser-tests
  ...
  hs-source-dirs: tests/
  main-is: ParserTests.hs

test-suite bios-tests
  ...
  hs-source-dirs: tests/
  main-is: BiosTests.hs

Using the documentation of cabal, we extract the four component names of the hie-bios project:

  • lib:hie-bios
  • exe:hie-bios
  • test:bios-tests
  • test:parser-tests

Since you rarely only want to load a single component in a cabal project, we have special syntax to be able to conveniently specify which directory belongs to which component. It is basically a multi-cradle.

cradle:
  cabal:
    - path: "./src"
      component: "lib:hie-bios"
    - path: "./exe"
      component: "exe:hie-bios"
    - path: "./tests/BiosTests.hs"
      component: "test:hie-bios"
    - path: "./tests/ParserTests.hs"
      component: "test:parser-tests"

Here you can see two important features:

  • We provide a mapping from filepath to component.
    • That way, we specify that a file such as ./src/HIE/Bios.hs belongs to the component lib:hie-bios.
  • The filepath can be a file.
    • This is convenient if components are overlapping.

This way we specified which component needs to be compiled given a certain source file for our whole project.

Debugging a cabal cradle

If you find that hie-bios can't load a certain component or file, you may run cabal repl <filename> and cabal repl <component name> to see if cabal succeeds in building the components. Chances are that there is a problem and if you fix that, hie-bios will succeed to load the project.

Also, see notes for testing your configuration.

Otherwise, please open an issue.

Ignoring directories

You can combine the multi-cradle with a none-cradle to ignore all source files in a certain directory. The syntax is a bit verbose:

cradle:
  multi:
    - path: "./tests/projects"
      config:
        cradle:
          none:
    - path: "./"
      config:
        cradle:
          cabal:
            - path: "./src"
              component: "lib:hie-bios"
            - path: "./exe"
              component: "exe:hie-bios"
            - path: "./tests/BiosTests.hs"
              component: "test:hie-bios"
            - path: "./tests/ParserTests.hs"
              component: "test:parser-tests"

This way, we specify that we do not want to load any files in our test project directories.

Bios

Alternatively you can explicitly state a program or shell command which should be used to collect the options. This is the most general approach and can be extended to handle arbitrary build systems.

The path of the program attribute is interpreted relative to the current working directory if it isn't absolute. A program is passed the file to return options for as its first argument, and a shell command will have it available in the HIE_BIOS_ARG environment variable.

There are two important environment variables:

  • HIE_BIOS_OUTPUT: describes the filepath the options should be written to. If this file does not exist, the program should create it.
  • HIE_BIOS_ARG: the source file that we want to load. Options returned by the program should be able to compile the given source file.
    • This environment variable is only available if a shell program is given.

The program flow is roughly as follows: The process must consult the HIE_BIOS_OUTPUT environment variable and write a list of options to this file, separated by newlines. Once the process finishes running, hie-bios reads this file and uses the arguments to set up the GHC session. This is how GHC's build system is able to support hie-bios. Note, the program is intended to produce the build flags to compile the whole component the given source file belongs to. This entails that the program lists all of the component's module- and file targets.

A good guiding specification for this file is that the following commands should work for any file in your project.

$ export HIE_BIOS_OUTPUT=./options.txt # this is usually some temporary file
$ ./<program> /path/to/foo.hs
$ ghci $(cat $HIE_BIOS_OUTPUT | tr '\n' ' ')

where HIE_BIOS_OUTPUT is some chosen output file and HIE_BIOS_ARG contains the file parameter.

The hie.yaml configuration looks like this:

cradle:
  bios:
    program: "<program>"

Alternatively, you may specify shell code directly. This is helpful, if your program executable consists of only a single call to another executable.

cradle:
  bios:
    shell: "<build-tool flags $HIE_BIOS_ARG>"

Additionally, you may specify the path to ghc. Otherwise, the one in the PATH will be used:

cradle:
  bios:
    program: "<program>"
    with-ghc: "<ghc>"

Debugging a bios cradle

The most common error in creating bios cradle is to not list all targets of the component. Please make sure, that you always list all targets of the component, associated with the filepath you want to load.

Also, see notes for testing your configuration.

Direct

The direct cradle allows you to specify exactly the GHC options that should be used to load a project. This is good for debugging but not a very good approach in general as the set of options will quickly get out of sync with a cabal file.

cradle:
  direct:
    arguments: [arg1, arg2]

Debugging a direct cradle

The arguments of a direct cradle will be passed almost directly to ghc. If the command ghc <cradle arguments> succeeds, then hie-bios can load the project.

None

The none cradle says that the IDE shouldn't even try to load the project. It is most useful when combined with the multi-cradle which is specified in the next section.

cradle:
  none:

Multi-Cradle

For a multi-component project you can use the multi-cradle to specify how each subdirectory of the project should be handled by the IDE.

The multi-cradle is a list of relative paths and cradle configurations. The path is relative to the configuration file and specifies the scope of the cradle. For example, this configuration specifies that files in the src subdirectory should be handled with the lib:hie-bios component and files in the test directory using the test:bios-tests component.

cradle:
  multi:
    - path: "./src"
      config:
        cradle:
          cabal:
            component: "lib:hie-bios"
    - path: "./test"
      config:
        cradle:
          cabal:
            component: "test:bios-tests"

If a file matches multiple prefixes, the most specific one is chosen. Once a prefix is matched, the selected cradle is used to find the options. This is usually a specific cradle such as cabal or stack but it could be another multi-cradle, in which case, matching works in exactly the same way until a specific cradle is chosen.

This cradle type is experimental and may not be supported correctly by some libraries which use hie-bios. It requires some additional care to correctly manage multiple components.

Note: Remember you can use the multi-cradle to declare that certain directories shouldn't be loaded by an IDE, in conjunction with the none cradle.

cradle:
  multi:
    - path: "./src"
      config: { cradle: {cabal: {component: "lib:hie-bios"}} }
    - path: "./test"
      config: { cradle: {cabal: {component: "test:bios-tests"}} }
    - path: "./test/test-files"
      config: { cradle: { none: } }

For cabal and stack projects there is a shorthand to specify how to load each component.

cradle:
  cabal:
    - path: "./src"
      component: "lib:hie-bios"
    - path: "./test"
      component: "test:bios-tests"
cradle:
  stack:
    - path: "./src"
      component: "hie-bios:lib"
    - path: "./test"
      component: "hie-bios:test:bios-tests"

Remember you can combine this shorthand with more complicated configurations as well.

cradle:
  multi:
    - path: "./test/testdata"
      config: { cradle: { none:  } }
    - path: "./"
      config: { cradle: { cabal:
                            [ { path: "./src", component: "lib:hie-bios" }
                            , { path: "./tests", component: "parser-tests" } ] } }

Cradle Dependencies

Sometimes it is necessary to reload a component, for example when a package dependency is added to the project. Each type of cradle defines a list of files that might cause an existing cradle to no longer provide accurate diagnostics if changed. These are expected to be relative to the root of the cradle.

This makes it possible to watch for changes to these files and reload the cradle appropiately. However, if there are files that are not covered by the cradle dependency resolution, you can add these files explicitly to hie.yaml. The file dependencies are not required to actually exist, since it can be useful to know when they are created, e.g. if there was no cabal.project in the project before and now there is, it might change how a file in the project is compiled.

Here's an example of how you would add cradle dependencies that may not be covered by the cabal cradle.

cradle:
  cabal:
    component: "lib:hie-bios"

dependencies:
  - package.yaml
  - shell.nix
  - default.nix

For the Bios cradle type, there is an optional field for specifying a program or shell command to obtain cradle dependencies from:

cradle:
  bios:
    program: ./flags.sh
    dependency-program: ./dependency.sh
cradle:
  bios:
    shell: build-tool flags $HIE_BIOS_ARG
    dependency-shell: build-tool dependencies

The dependency program or command is executed with no parameters and it is expected to output on stdout on each line exactly one filepath relative to the root of the cradle, not relative to the location of the program.

Programs and shell commands for flags and dependencies can be mixed and matched.

Configuration specification

The complete configuration is a subset of

cradle:
  cabal:
    component: "optional component name"
  stack:
    component: "optional component name"
  bios:
    program: "program to run"
    dependency-program: "optional program to run"
    shell: build-tool flags $HIE_BIOS_ARG
    dependency-shell: build-tool dependencies
    with-ghc: "optional path to ghc"
  direct:
    arguments: ["list","of","ghc","arguments"]
  none:
  multi: - path: ./
           config: { cradle: ... }

dependencies:
  - someDep

Testing your configuration

The given hie-bios executable is provided to test your configuration.

The flags command will print out the options that hie-bios thinks you will need to load a file.

$ hie-bios flags exe/Main.hs

The check command will try to use these flags to load the module into the GHC API.

$ hie-bios check exe/Main.hs

The debug command prints verbose information about the cradle, such as where the hie.yaml file was found, which file is loaded and the options that will eventually be used for loading a session.

$ hie-bios debug exe/Main.hs

Implicit Configuration

There are several built in modes which capture the most common Haskell development scenarios. If no hie.yaml configuration file is found then an implicit configuration is searched for. It is strongly recommended to just explicitly configure your project.

Priority

The targets are searched for in the following order.

  1. A specific .hie-bios file.
  2. A stack project
  3. A cabal project
  4. The direct cradle which has no specific options.

cabal-install

The workspace root is the first folder containing a cabal.project file.

The arguments are collected by running cabal v2-repl <filename>.

If cabal v2-repl <filename> fails, then the user needs to configure the correct target to use by writing a hie.yaml file.

stack

The workspace root is the first folder containing a stack.yaml file.

The arguments are collected by executing stack repl. If this fails, the user needs to configure the correct target to use by writing a hie.yaml file.

bios

The most general form is the bios mode which allows users to specify which flags to provide themselves.

The program will receive the file to return options for as its first argument.

The program flow is roughly as follows: The process must consult the HIE_BIOS_OUTPUT environment variable and write a list of options to the file pointed to by HIE_BIOS_OUTPUT, separated by newlines. Once the process finishes running, hie-bios reads this file and uses the arguments to set up the GHC session. This is how GHC's build system is able to support hie-bios. Note, the program is intended to produce the build flags to compile the whole component the given source file belongs to. This entails that the program lists all of the component's module- and file targets.

A good guiding specification for this file is that the following commands should work for any file in your project.

$ export HIE_BIOS_OUTPUT=./options.txt # this is usually some temporary file
$ ./.hie-bios /path/to/foo.hs
$ ghci $(cat $HIE_BIOS_OUTPUT | tr '\n' ' ')

This is useful if you are designing a new build system or the other modes fail to setup the correct session for some reason. For example, this is how hadrian (GHC's build system) is integrated into hie-bios.

Supporting Bazel and Obelisk

In previous versions of hie-bios there was also support for projects using rules_haskell and obelisk. This was removed in the 0.3 release as they were unused and broken. There is no conceptual barrier to adding back support but it requires a user of these two approaches to maintain them.