Moved the call to parseExprKeyPathObjC from parseExprUnary to parseExprPostfixWithoutSuffix. This allows to perform operations on #keyPath(...) as it now parses the suffix for it.
Fixes [Swift JIRA] (SR-5605) Parser fails on optional #keyPath in cas
Eg. `#keyPath(SomeClass.someProperty).description` is now possible
A string interpolation expression is composed of { OpenQuote, Segments,
CloseQuote }. To represent OpenQuote, CloseQuote and StringSegment, we have to
introduce new token kinds correspondingly.
This patch also refactors the structure of function signature node so
that closure signature can re-use parts of function signature. For
instance, we group arrow and return type to be "ReturnClause". And we
group parenthesized parameter list to be "ParamClause".
This structure of closure signature also calls for a good way to
represent either-or node in libSyntax APIs, since we've two ways to
specify parameters in closure: one is as regular function parameter and
the other is dot-separated simple names.
* libSyntax: Parse member access expression.
This patch uses createNodeInPlace from syntax parsing context API to
merge an expression with its suffix to create recursive nodes such as
member access expression.
Meanwhile, this patch breaks down a signed integer or float literal to a
prefix operator expression. This expression consists of two parts: an
operator and the following expression. This makes literals like "+1" or
"-1" no different from other prefix unary expressions such as "!true".
Tuple expression essentially has the same underlying structure as
function call arguments in libSyntax. However, we separate them as
different libSyntax kinds for better usability.
Different from AST, libSyntax currently allows single-child,
label-free tuple expressions (represented as ParenExpr in AST). This is
subject to change if we need to adopt the same differentiation in
libSyntax in the future.
Along with starting to support ternary expressions, this commit also
slightly changes SyntaxParsingContext APIs as follows:
1. Previously, makeNode() only supports node creation by using the nodes
from the underlying syntax token array; this commit allows it to use the nodes from
the pending syntax list as well.
2. This commit strictly limits that the pending syntax list should never
contain token syntax node.
3. The node kind test shouldn't include unknown kinds. They are noisy.
