Can someone point me to best practices for error handling in a Visual Studio Code extension?
I'm writing an extension in TypeScript that contributes a debugger. I want to log unexpected behavior, sometimes as information to the user explaining that something didn't go right, sometimes to create a trail for debugging, but certainly not to fail silently. Using console.log or console.error shows up in the debug output when I am debugging the extension, but I can't find it when the extension is installed. Do I have to open an output channel specifically for my extension and write everything there? Should I be throwing up showInformationMessage and showErrorMessage windows? Should I just be throwing exceptions and hope that code will do the right thing?
In my extension I use two ways for feedback. One is an output channel for errors produced by an external process that I'm using for the work.
Create the channel in your activation method:
outputChannel = window.createOutputChannel("ANTLR4 Errors");
and push output to it whenever you have something:
} catch (reason) {
outputChannel.appendLine("Cannot parse sentence generation config file:");
outputChannel.appendLine((reason as SyntaxError).message);
outputChannel.show(true);
return;
}
The other one is what you already considered:
if (workspace.getConfiguration("antlr4.generation").mode === "none") {
void window.showErrorMessage("Interpreter data generation is disabled in the preferences (see " +
"'antlr4.generation'). Set this at least to 'internal' to enable debugging.");
return null;
}
which is for messages I create in the extension and that users need to see and take seriously.
Related
I am writing a VS Code extension that depends on the currently set Python interpreter. When I change the Python Interpreter via the VS Code UI, the extension needs to refresh and get the latest Python path (mainly to show the right environment settings in the TreeView). For now, I have a refresh button in my custom TreeView that I need to press after selecting a different Python interpreter.
However, this is a second manual step. Is there a way to get a notification in my extension, when a user changes the Python Interpreter, e.g., an event the extension can listen to?
I only found VS Code's Activation Events, but it doesn't look like this would help. I didn't find any other events that get triggered after the command python.setInterpreter is executed
Finally found it. The right config to watch for is python.defaultInterpreterPath
vscode.workspace.onDidChangeConfiguration(event => {
let affected = event.affectsConfiguration("python.defaultInterpreterPath");
if (affected) {
doSomething();
}
});
To support the usingNewInterpreterStorage case (default today), add:
const extension = vscode.extensions.getExtension('ms-python.python')!;
await extension.activate();
extension.exports.settings.onDidChangeExecutionDetails((event: any) => {
doSomething();
});
I am working on a VsCode extension in that I want to provide custom snippets for code completion.
I know about the option of using snippet json files directly, however those have the limitation of not being able to utilize the CompletionItemKind property that determines the icon next to the completion suggestion in the pop-up.
My issue:
If I implement a simple CompletionItemProvider like this:
context.subscriptions.push(
vscode.languages.registerCompletionItemProvider(
{scheme:"file",language:"MyLang"},
{
provideCompletionItems(document: vscode.TextDocument, position: vscode.Position) {
let item = new vscode.CompletionItem('test');
item.documentation = 'my test function';
item.kind = vscode.CompletionItemKind.Function;
return [item];
}
}
)
)
then the original VsCode IntelliSense text suggestions are not shown anymore, only my own. Should I just return a kind of an empty response, like
provideCompletionItems(document: vscode.TextDocument, position: vscode.Position) {
return [null|[]|undefined];
}
the suggestions appear again as they should. It seems to me that instead of merging the results of the built-in IntelliSense and my own provider, the built-in ones get simply overridden.
Question:
How can I keep the built-in IntelliSense suggestions while applying my own CompletionItems?
VsCode Version: v1.68.1 Ubuntu
I seem to have found the answer for my problem, so I will answer my question.
Multiple providers can be registered for a language. In that case providers are sorted
by their {#link languages.match score} and groups of equal score are sequentially asked for
completion items. The process stops when one or many providers of a group return a
result.
My provider seems to provide results that are just higher scored than those of IntelliSense.
Since I didn't provide any trigger characters, my CompletionItems were comteping directly with the words found by the built-in system by every single pressed key and won.My solution is to simply parse and register the words in my TextDocument myself and extend my provider results by them. I could probably just as well create and register a new CompletionItemProvider for them if I wanted to, however I decided to have a different structure for my project.
In my VS Code extension I have some code that uses the built in Markdown extension. I capture a reference to it by registering as a markdown plugin and putting the following code at the end of my extension's activate method.
return {
extendMarkdownIt(mdparam: any) {
return md = mdparam;
}
};
Markdown calls this when it activates.
Generally this is not a problem. Most of the use cases for my extension involve a markdown file already loaded into the active editor, and the loading of this file triggers activation of the markdown extension.
However there are some legitimate use cases in which this is not so.
I need to programmatically trigger activation of the markdown extension. Some of these cases involve having a different kind of file open in the active editor so loading a markdown file into it is not an acceptable option.
Some potential strategies:
Change the language mode. There is a command workbench.action.editor.changeLanguageMode but no documentation. I tried
vscode.commands.executeCommand('workbench.action.editor.changeLanguageMode', 'md');
but this triggers the UI
so I tried a pattern I've seen in the parameters of other commands and added , true. This suppressed the UI but doesn't seem to work.
Load a markdown file into a new editor then close it again. This should work, but it's ugly.
Put something in the contributions section of my extension that changes the activation trigger for the markdown extension so that it is triggered by the other file types on which my extension operates.
Of these options my favourite would be 3 but I don't even know whether this is even possible. Option 1 is hampered by the crappy (in many cases non-existent) documentation for vscode internal commands.
Option 1 it is. If anyone knows how to do option 3 please tell, the solution below is a ghastly hack.
It is possible to trigger activation of the Markdown extension by changing the document language of any open editor to markdown. In the event that there are no open editors a document with the markdown language set can be created in memory and loaded into an editor.
If VS Code is busy loading extensions activation can take several hundred milliseconds so the best thing to do is watch the variable into which markdown-it is captured.
The variable md is a global (global to my extension, not the whole of VS Code) into which a reference is acquired as shown in the question.
let ed = vscode.window.activeTextEditor;
if (ed) {
let lid = ed.document.languageId;
if (lid !== "markdown") {
vscode.languages.setTextDocumentLanguage(ed.document, "markdown").then(
function waitForMd() {
if (md) {
vscode.languages.setTextDocumentLanguage(ed!.document, lid);
} else {
setTimeout(waitForMd, 100);
}
}
);
}
} else {
vscode.workspace.openTextDocument({ language: "markdown" }).then(doc => {
vscode.window.showTextDocument(doc).then(
function waitForMd() {
if (md) {
vscode.commands.executeCommand("workbench.action.closeActiveEditor");
} else {
setTimeout(waitForMd, 100);
}
});
});
}
Once the capture completes we can restore the true language or close the editor as appropriate. To be realistic the second case (no active editor) is unlikely because my own extension won't activate until you load something. At any rate it works stably now. The larger project is progressing nicely.
I love testing-library, have used it a lot in a React project, and I'm trying to use it in an Angular project now - but I've always struggled with the enormous error output, including the HTML text of the render. Not only is this not usually helpful (I couldn't find an element, here's the HTML where it isn't); but it gets truncated, often before the interesting line if you're running in debug mode.
I simply added it as a library alongside the standard Angular Karma+Jasmine setup.
I'm sure you could say the components I'm testing are too large if the HTML output causes my console window to spool for ages, but I have a lot of integration tests in Protractor, and they are SO SLOW :(.
I would say the best solution would be to use the configure method and pass a custom function for getElementError which does what you want.
You can read about configuration here: https://testing-library.com/docs/dom-testing-library/api-configuration
An example of this might look like:
configure({
getElementError: (message: string, container) => {
const error = new Error(message);
error.name = 'TestingLibraryElementError';
error.stack = null;
return error;
},
});
You can then put this in any single test file or use Jest's setupFiles or setupFilesAfterEnv config options to have it run globally.
I am assuming you running jest with rtl in your project.
I personally wouldn't turn it off as it's there to help us, but everyone has a way so if you have your reasons, then fair enough.
1. If you want to disable errors for a specific test, you can mock the console.error.
it('disable error example', () => {
const errorObject = console.error; //store the state of the object
console.error = jest.fn(); // mock the object
// code
//assertion (expect)
console.error = errorObject; // assign it back so you can use it in the next test
});
2. If you want to silence it for all the test, you could use the jest --silent CLI option. Check the docs
The above might even disable the DOM printing that is done by rtl, I am not sure as I haven't tried this, but if you look at the docs I linked, it says
"Prevent tests from printing messages through the console."
Now you almost certainly have everything disabled except the DOM recommendations if the above doesn't work. On that case you might look into react-testing-library's source code and find out what is used for those print statements. Is it a console.log? is it a console.warn? When you got that, just mock it out like option 1 above.
UPDATE
After some digging, I found out that all testing-library DOM printing is built on prettyDOM();
While prettyDOM() can't be disabled you can limit the number of lines to 0, and that would just give you the error message and three dots ... below the message.
Here is an example printout, I messed around with:
TestingLibraryElementError: Unable to find an element with the text: Hello ther. This could be because the text is broken up by multiple elements. In this case, you can provide a function for your text matcher to make your matcher more flexible.
...
All you need to do is to pass in an environment variable before executing your test suite, so for example with an npm script it would look like:
DEBUG_PRINT_LIMIT=0 npm run test
Here is the doc
UPDATE 2:
As per the OP's FR on github this can also be achieved without injecting in a global variable to limit the PrettyDOM line output (in case if it's used elsewhere). The getElementError config option need to be changed:
dom-testing-library/src/config.js
// called when getBy* queries fail. (message, container) => Error
getElementError(message, container) {
const error = new Error(
[message, prettyDOM(container)].filter(Boolean).join('\n\n'),
)
error.name = 'TestingLibraryElementError'
return error
},
The callstack can also be removed
You can change how the message is built by setting the DOM testing library message building function with config. In my Angular project I added this to test.js:
configure({
getElementError: (message: string, container) => {
const error = new Error(message);
error.name = 'TestingLibraryElementError';
error.stack = null;
return error;
},
});
This was answered here: https://github.com/testing-library/dom-testing-library/issues/773 by https://github.com/wyze.
From what I can tell, the assembly containing the addin must be located in C:\Program Files (x86)\NUnit 2.5.7\bin\net-2.0\addins. I think my assembly is being loaded because I have to close NUnit-gui before I can replace the assembly in the addins directory. The problem, however, is that I don't see any of the effects of the addin (None of the event handlers are being called)
So how do I verify that my addin has been loaded? I'd love to step through with the debugger but I'd be perfectly happy with print line debugging. When I tried doing a File.WriteAllText() the addin failed to load but gave no reason. Also, how do I debug the loading process?
The NUnit docs are helpful, but they're bare bones at best when it comes to extensibility and there isn't any intellisense available for classes in NUnit.Core.
You should use some tracing library like this one which you can download here.
Now you can decorate your relevant methods with using statements like this:
using ApiChange.Infrastructure;
class MyAddin
{
static TypeHashes myType = new TypeHashes(typeof(MyAddin);
void RelevantMethod()
{
using (Tracer t = new Tracer(myType, "RelevantMethod"))
{
....
if(bLoaded == false)
t.Error("Could not load adding because of {0}", reason);
}
}
}
Then you can enable tracing via the environment variable _TRACE
set _Trace=debugoutput
DebugOutput can be viewed with the SysInternals tool DbgView (no attach simply start it and watch the traces).
Or you trace to a file
set _Trace=file
The trace file is located where the executable is e.g. Nunit.exe.txt. If you set _TRACE to some random string it will trace the help to console and OutputDebugString to give you help.
Why this trace library? It is actually the ONLY one which is able to trace any exception when your method is left. This does work when the method contains using statements for tracing like the one above. If it is actually your fault that NUnit does choose to ignore your plugin you can find out now.
The output will look like this:
* ApiChange.IntegrationTests.Diagnostics.TracingTests.Demo_Show_Leaving_Trace_With_Exception
18:57:46.665 03064/05180 <{{ > ApiChange.IntegrationTests.Diagnostics.TracingTests.SomeMethod
18:57:46.668 03064/05180 <{{ > ApiChange.IntegrationTests.Diagnostics.TracingTests.SomeOtherMethod
18:57:46.670 03064/05180 < }}< ApiChange.IntegrationTests.Diagnostics.TracingTests.SomeOtherMethod Exception thrown: System.NotImplementedException: Hi this a fault
at ApiChange.IntegrationTests.Diagnostics.TracingTests.FaultyMethod()
at ApiChange.IntegrationTests.Diagnostics.TracingTests.SomeOtherMethod()
at ApiChange.IntegrationTests.Diagnostics.TracingTests.SomeMethod()
at ApiChange.IntegrationTests.Diagnostics.TracingTests.Demo_Show_Leaving_Trace_With_Exception()
18:57:46.670 03064/05180 < }}< ApiChange.IntegrationTests.Diagnostics.TracingTests.SomeOtherMethod Duration 2ms
18:57:46.689 03064/05180 < }}< ApiChange.IntegrationTests.Diagnostics.TracingTests.SomeMethod Duration 24ms
That should make it easy to find out why your Addin was not used at all. And you do not need a debugger ;-).
Yours,
Alois Kraus