lgolding / auto-scale Goto Github PK

The language service seems to reuse the same TokenInfo object each time it calls the Scanner. So be sure to set the Trigger property to TokenTriggers.None before OR-ing in the appropriate triggers, if any, for the current token.

Failing to do this meant that the triggers were accumulated from call to call. So, for example, once a ParenClose token was seen, all subsequent tokens had the TokenTriggers.MatchBraces flag set. This caused the language service to do more work than it needed to (attempting to match braces after every token).

Repro:

^ = 1;\n

This should produce one error (ASF0001 (0-0): Expected a token of type Identifier, but got token "^" of type Unknown.). But it produces an additional, spurious error (ASF0001 (7-7): Expected a token of type Identifier, but got token "" of type End.).

This does not happen if the trailing newline is omitted, and it does not happen if the assignment statement is correct (a = 1).

For example, this is illegal:

$ActiveTasks = 3;

(This is also illegal, but it's caught as a parse error because we only allow an identifier on the left-hand side of an assignment:

$ActiveTasks.Count = 3;

)

Rules generally only look at one kind of node. Don't call all of their Visit methods, just the ones they override.

When a string like 1.9.0 occurs within a function argument list, for example:

a = $ActiveTasks.GetSample(1.9.0)

it gives a parse error at the second dot, which is fine. But the error message is

Expected a token of one of the types DoubleLiteral, StringLiteral, Identifier,
but got token "." of type OperatorMemberSelect.

Why would the parser expect on of those at the point where I typed the second dot?

Example: if you type a . at the end of this line (which already has a . at the end):

$ActiveTasks.

... it displays the member selection list for $ActiveTasks, even though there is an intervening token.

It should only display the member selection list if the identifier is the immediately preceding token.

Count comments as white space in Lexer.SkipWhite.

... because I labeled the constants taskcompletion, etc., as Keyword, and the PrimaryExpression grammar doesn't allow Keyword.

All the items in the Intellisense lists use icon 0. Use different icons for user-defined variables, methods, keywords (taking hints from what C# lists look like).

If you type a . between the identifiers $ActiveTasks and GetSample on this line:

$ActiveTasksGetSample

you don't get a member selection list. That's because the code incorrectly considers the "preceding" identifier to be GetSample (which has no members), rather than $ActiveTasks (which does have members).

If you leave a method invocation unclosed, for example

a = $ActiveTasks.GetSample(1.9

the parser attempts to construct a MethodInvocationNode with the closeParen argument set to null. This gives a NullReferenceException when the argument closeParen.EndIndex is passed to the base class ctor.

If there is no closing paren, throw a parse exception instead of trying to construct the MethodInvocationNode.

The test methods in those classes all have the same structure.

We can't put the entire test method into a base class, because the test method is decorated with [MemberData("TestCases")], and Xunit insists that a member of that name exist on the base class. The best we can do is factor the bodies of the test methods into a common helper in the base class.

Translate symbol names like OpenParen into English in the messages.

Shorter and cleaner.

Example: myVar.GetSample() is invalid. So is $NodeDeallocationOption.GetSample().