Skip to content

Latest commit

 

History

History
277 lines (211 loc) · 10.6 KB

HACKING.md

File metadata and controls

277 lines (211 loc) · 10.6 KB

Developing on the Ada Language Server

Dependencies

The Ada Language Server depends on a number of tools and libraries listed on the main page.

Coding

The ALS repository includes a template VS Code workspace at .vscode/settings.json.tmpl. This allows developers to instantiate the workspace locally and customize it to their liking, in particular the terminal.integrated.env.<os> setting to set PATH and GPR_PROJECT_PATH as needed to provided tools and dependencies to the development environment.

To instantiate the VS Code workspace, simply run make configure. Then open VS Code at the root of the ALS repository.

At the first run, the workspace will recommend installing a set of extensions including a released version of the Language Support for Ada extension which is produced from this repository. That allows you to work with the Ada codebase of the ALS.

Building

To build the ALS, run:

make

Debugging

One way to run the ALS is by running the VS Code extension in development mode. You can do that using the VS Code workspace at the root of this repository and the launch configuration (vscode) Launch vscode with the Ada extension.

You can activate traces that show all the server input/output. This is done by creating a file $HOME/.als/traces.cfg with the following contents:

ALS.IN=yes > inout.txt:buffer_size=0
ALS.OUT=yes > inout.txt:buffer_size=0

When this is present, the Ada Language Server will generate a file $HOME/.als/inout.txt which logs the input received and the output sent by the language server.

You can also monitor the server's memory usage by adding the following lines:

>ada_ls_log.$T.txt:buffer_size=0:buffer_size=0
ALS.MAIN=yes
DEBUG.ADA_MEMORY=yes

This will create ada_ls_log.<timestamp>.txt log files in your $HOME/.als directory each time the Ada Language Server is run, with additional information about memory usage, allowing to track down which parts of the program consumes the most memory. This is very useful to resolve unexpected memory consumption issues.

Getting symbolic backtrace

To get a symbolic backtrace you need the debug information files. For the release you can download them from GitHub Release Assets. Assets contain an archive per platform with the ALS executable and debug information files (.debug file on Linux/Windows and .dSYM directory for Mac OS X). Extract the debug information to the directory containing ALS (usually this is $HOME/.vscode/extensions/adacore.ada-*/{darwin,linux,win32}) and run

addr2line ada_language_server <hex backtrace>

On Mac OX X use atos instead.

Getting symbolic backtrace for TypeScript

The Ada & SPARK extension for Visual Studio Code is published as minified JavaScript code. For that reason backtraces reported are not easily readable by humans. To alleviate that, the published .vsix includes a source map file at out/src/extension.js.map.

To resolve a backtrace to the original TypeScript source locations:

  1. Unzip the .vsix

  2. Copy the backtrace into a plain text file while preserving the error message on the first line, for example:

    2024-01-11 10:11:54.191 [Ada Extension] ERROR    Error while starting Ada extension...
        at Gu (/home/XXXXXXX/.vscode/extensions/adacore.ada-24.0.3/out/src/extension.js:75:5911)
        at oS (/home/XXXXXXX/.vscode/extensions/adacore.ada-24.0.3/out/src/extension.js:75:6057)
        at w$ (/home/XXXXXXX/.vscode/extensions/adacore.ada-24.0.3/out/src/extension.js:85:2292)
        at b$ (/home/XXXXXXX/.vscode/extensions/adacore.ada-24.0.3/out/src/extension.js:85:2154)
        ...
    
  3. Run the following command to obtain source locations:

    $ npx stacktracify out/src/extension.js.map -f trace.txt
    2024-01-11 10:11:54.191 [Ada Extension] ERROR    Error while starting Ada extension...
        at [unknown] (../../src/helpers.ts:150:56)
        at getEvaluatedCustomEnv (../../src/helpers.ts:186:22)
        at assertSupportedEnvironments (../../src/extension.ts:128:4)
        at activateExtension (../../src/extension.ts:118:14)
    

Writing tests

To write a functional test for Ada Language Server:

  • Choose a meaninful name for your test, for instance completion_inside_generics. We'll refer to this as <testname> below
  • Create a new directory testsuite/ada_lsp/<testname> containing your test data
  • Activate full in/out language server traces. See the Debugging section above.
  • From the base directory in this repository, run:
    python scripts/traces_to_test.py <testname> <path_to_the_als_traces_file>
    
  • Delete all extra requests, notifications and capabilities that are not related to the feature being tested. It will help a lot later when tests are baselined because of changes in protocol, capablitities or message formats.
  • Replace the comment at the beginning of the test with something meaningful:
    "comment": [
       "test automatically generated"
    ]

Running tests

  • run make check to run the entire testsuite
  • to run an individual test, go to testsuite and run sh run.sh ada_lsp/<testname> (you will need https://github.com/AdaCore/e3-testsuite installed to do this)

VS Code integration tests

Run make vscode-test to run the VS Code testsuite.

If you open the ALS repository in VS Code, it is also possible to run VS Code integration tests using the Testing view. The integration/vscode/ada/.vscode-test.mjs contains a configuration allowing to load the tests in the Testing view. The UI offers ways to run or debug one test, a subset of tests or all tests.

Other tests

See more about the project testsuite here.

Release VS Code extension

To release a new version of the VS Code extension, just put a tag with of the form 22.0.3, where 22.0 matches the current GNAT Studio development version. The corresponding git command is

   git checkout edge
   git tag -a 22.0.3

Put release notes in the comment of the tag starting with a Release notes header:

Release notes

* <notes 1>
* <notes 2>, etc.

Then push it to GitHub with git push 22.0.3 or git push --tags. The GitHub CI will publish the extension on the marketplace and the Open VSX Registry. It also makes GitHub Release and put archives with ALS executables and debug information files.

Source directories

The repository contains the following folders:

  • doc/ - documentation of the project and LSP extensions
  • gnat/ - GNAT project files
  • integration/vscode/ada/ - VS Code extension sources
  • scripts/ - some support scripts
  • ada_language_server/source/ada/ - LSP implementation for Ada language
  • source/client/ - basic client interface
  • source/protocol/ - LSP binding in Ada
  • source/server/ - Common LSP server implementation
  • source/spawn/ - Process spawn/communication API
  • source/tester/ - source of the test driver
  • testsuite/ada_lsp/ - test suite of LSP in form of request/response
  • subprojects/ - where we store dependencies (VCS, ada_libfswatch)
  • subprojects/stubs - the location for "stub" versions of .gpr files, for dependencies that are optional.

Protocol synchronization

You can get the latest Language Server Protocol specification from the https://github.com/microsoft/language-server-protocol.git repository, gh-pages branch.

Currently, we synchronize Ada sources with LSP updates manually. The Ada source to synchronize is the LSP.Messages package. The specification of the package contains TypeScript snippets from the Language Server Protocol specification in the form of comments:

   --```typescript
   --interface Message {
   --   jsonrpc: string;
   --}
   --```

You can easily extract all such snippets from the specification with the command

sed -n -e '/^```typescript/,/^```/p' specification-3-14.md |sed -e 's/^/   --/'

To extract corresponding snippets from lsp-messages.ads use this:

sed -n -e '/^   --```typescript/,/^   --```/p' lsp-messages.ads

Unfortunately we have to reorder Ada declarations to follow "define before use" rule. To restore original order use scripts/reorder.py script. Then you can compare texts with diff -u to see what's changed.

Rules to match method name with corresponding messages are not present in the TypeScript snippets. So we don't put requests, responses and notifications types in then LSP.Messages package. Instead we have 6 packages divided by message kind and directions:

  • LSP.Messages.Server_Requests - requests from the client to the server
  • LSP.Messages.Server_Notifications - notifications from the client to the server
  • LSP.Messages.Server_Responses - responses from the server to the client
  • LSP.Messages.Client_Requests - requests from the server to the client
  • LSP.Messages.Client_Notifications - notifications from the server to the client
  • LSP.Messages.Client_Responses - responses from the client to the server

Server_Requests and Server_Notifications packages are generated by the scripts/generate.py script.

Each of these 6 packages defines common root type and a Visit subprogram to implement Visitor pattern. Corresponding visitors are defined in packages:

  • LSP.Server_Request_Receivers
  • LSP.Server_Notification_Receivers
  • Server_Response_Sender
  • LSP.Client_Request_Receivers
  • LSP.Client_Notification_Receivers
  • LSP.Client_Response_Senders

The type Message_Logger implements all of these to provide custom format for each message kind in the trace.

The LSP server uses its-own Server_Request_Handler interface. This interface defines functions corresponding to client-to-server LSP requests. Function profiles ensure correct argument and result types at compile time.

LSP.Ada_Handlers implements Server_Request_Handler and Server_Notification_Receiver interfaces to provide Ada language support. It leverages Server to send notification to the client.

Error_Decorator also implements Server_Request_Handler and provides safety net to catch Property_Error exceptions during request processing.

LSP Server, on its side, converts messages to/from JSON, tracks request Ids and implements the CancelRequest logic.

The LSP.Ada_Documents.Document type represents a file opened in an editor. It keeps text of a changed file.

The LSP.Ada_Contexts.Context type represents non-aggregate project tree and keeps a Libadalang context to process compilation units from this tree.

Document and Context objects belong to Ada_Handler.