This reverts the changes
commit beac7b9539
C++: Fix highlighting after "invalid code"
commit 78ab287fc6
C++: Stop parsing a declaration after two tries
which were a work around for QTCREATORBUG-12890.
A follow-up patch provides a proper fix.
Task-number: QTCREATORBUG-12890
Change-Id: I2650a8e41c8ff1180cad9f069e463fc51bd2f1b1
Reviewed-by: Erik Verbruggen <erik.verbruggen@theqtcompany.com>
For the semantic info document we do not expand function like macros and
because of that certain macro invocations lead to invalid code that we
need to handle, e.g.:
Q_GLOBAL_STATIC(CppTools::SymbolFinder, symbolFinder)
class Foo {};
This change makes parsing Foo in the semantic info document successfully
again, which affects highlighting of that class.
Change-Id: I389265ac64d3f0b8b8f406d38fa58d78820b14ba
Reviewed-by: Erik Verbruggen <erik.verbruggen@theqtcompany.com>
If we fail to parse a declaration, we rewind, eat the token and look for the
next token that might be a good candidate for a declaration start (e.g. an
identifier). This becomes cpu and memory expensive with super long and invalid
expressions like
typedef b:Ⓜ️:if_< b:Ⓜ️:bool_<
(sizeof(fun((Dummy *) 0, (ThisT *) 0, (b:Ⓜ️:int_<70> *) 0)) ==
sizeof(defined_)) >, b:Ⓜ️:if_< b:Ⓜ️:bool_<
(sizeof(fun((Dummy *) 0, (ThisT *) 0, (b:Ⓜ️:int_<71> *) 0)) ==
sizeof(defined_)) >, b:Ⓜ️:if_< b:Ⓜ️:bool_<
(sizeof(fun((Dummy *) 0, (ThisT *) 0, (b:Ⓜ️:int_<72> *) 0)) ==
sizeof(defined_)) >, b:Ⓜ️:if_< b:Ⓜ️:bool_<
(sizeof(fun((Dummy *) 0, (ThisT *) 0, (b:Ⓜ️:int_<73> *) 0)) ==
sizeof(defined_)) >, b:Ⓜ️:if_< b:Ⓜ️:bool_<
(sizeof(fun((Dummy *) 0, (ThisT *) 0, (b:Ⓜ️:int_<74> *) 0)) ==
sizeof(defined_)) >, b:Ⓜ️:if_< b:Ⓜ️:bool_<
(sizeof(fun((Dummy *) 0, (ThisT *) 0, (b:Ⓜ️:int_<75> *) 0)) ==
sizeof(defined_)) >, b:Ⓜ️:if_< b:Ⓜ️:bool_<
// ...some more crazy lines like this
Therefore, stop trying after two failures by looking for the next semicolon or
closing curly brace.
Task-number: QTCREATORBUG-12890
Change-Id: I6637daeb840dd549d669080775228fa91fc932eb
Reviewed-by: Erik Verbruggen <erik.verbruggen@theqtcompany.com>
It should be parsed as an DeclarationStatement, but instead it was
parsed as an ExpressionStatement.
Regression introduced with
commit d3c5fff66d.
C++: Fix expensive parsing of expressions
The introduced ASTCache did not save the correct return value of a
parse* function. Because of that, the first return in
Parser::parseExpressionList returned false on the second invocation
(cache hit), instead of true, which resulted in an ExpressionStatement.
Task-number: QTCREATORBUG-13122
Change-Id: I8dbd8852b0909edddcd3195b484f4cea92328cc5
Reviewed-by: Fawzi Mohamed <fawzi.mohamed@digia.com>
For expression statements like "(g(g(g(...(g(0))...))))" we reparsed
quite much again and again for nothing. The high-level trace for this
expression looks like this:
parseCastExpression
parseTypeId
parseAbstractDeclarator
parseAbstractCoreDeclarator
parseParameterDeclarationClause (--> DEEP)
...
parseUnaryExpression
...
parseCorePostfixExpression
parseTypeId (--> DEEP)
parsePrimaryExpression (--> DEEP)
Especially parseTypeId is expensive in this case and it's called two
times, both from the same token (index).
With this patch, we remember for certain ASTs the parse results and
re-use them when needed.
Change-Id: I013d1c064c655636bc94db408097863b5e183fc2
Task-number: QTCREATORBUG-12252
Reviewed-by: Erik Verbruggen <erik.verbruggen@digia.com>
A case or a default statement must be followed by another statement.
When a such a case (or default) statement is followed immediately by
another case (or default) statement, then this would create a linked
list, and the parser will recurse to parse such input.
In order to prevent the parser running out of stack space while
recursing, parse this corner case by blocking parsing a labeled
statement as the first statement after a labeled statement.
The advantage is that these statements do not form a linked list, so any
subsequent visitation of the AST won't run out of stack space either.
Change-Id: Id2111a49509132997f5fbe4bb12c92c729ec2522
Task-number: QTCREATORBUG-12673
Reviewed-by: Nikolai Kosjar <nikolai.kosjar@digia.com>
The introduction of C99 designators led to parsing problems with lambdas
that were passed in as a function arguments.
Fixed by prefering to parse without designators first. This will be
cleaner/clearer once the appropriate "LanguageFeatures" from the Project
Parts will be passed in.
Change-Id: Ia9cb7c4a4c9345e729cf2044e1e5411fe63e33ec
Reviewed-by: Wang Hoi <wanghoi@126.com>
Reviewed-by: Erik Verbruggen <erik.verbruggen@digia.com>
The parser considers '{' an operator and thus thought
"template-id {" couldn't possibly be valid. This patch adds
'{' as an exception to the rule.
Change-Id: I40730fcdc5cade48566b4c8b6fde390f455bbdba
Reviewed-by: Nikolai Kosjar <nikolai.kosjar@digia.com>
As the name function name suggests, true should be returned if the
declaration should be processed. Otherwise false.
Change-Id: I8d266d99c579b331fee8772bde47aa1a466dae9c
Reviewed-by: Erik Verbruggen <erik.verbruggen@digia.com>
CPlusPlus::Parser::parseAccessSpecifier() is not used anywhere
throughout the QtCreator codebase.
Change-Id: I062a4ae257b9c61f02bf85079feb6d48bd07c49e
Reviewed-by: Erik Verbruggen <erik.verbruggen@digia.com>
This is an obvious typo considering the checks at the beginning of the function.
Change-Id: I11b784153b650b02bba7bb942ca1131447a3da4a
Reviewed-by: Erik Verbruggen <erik.verbruggen@digia.com>
MEMBER was added in Qt5.
Task-number: QTCREATORBUG-10068
Change-Id: Ic6c15a0e5ee8981ab98e4c12fc1521dc281b731f
Reviewed-by: Erik Verbruggen <erik.verbruggen@digia.com>
No semantic analysis yet, but this prevents the parser from generating
bogus diagnostics.
Task-number: QTCREATORBUG-9309
Change-Id: I2ec575a8474cd51bfa97b17678d3da71ab8dcd7a
Reviewed-by: Nikolai Kosjar <nikolai.kosjar@digia.com>
C++ standard defines that 'try' should be before optional ctor
initializer, so wrong order changed. Added documentation to
parseTryBlockStatement and new test.
Task-number: QTCREATORBUG-9064
Change-Id: Id19cdc53c034cb1232ae27e0bfe36d85b7ad0452
Reviewed-by: Erik Verbruggen <erik.verbruggen@digia.com>
When a constructor is defined with a single, unnamed argument of a custom type without
extra type specifiers (const...), then the constructor was not identified as such.
There was an heuristic in case the constructor was in the class definition, but not if the
the constructor was defined later.
Examples:
class Arg;
class Other;
class Foo {
Foo(Arg /*arg*/); // working
Foo(const Arg /*arg*/); // working
Foo(int /*arg*/); // working
Foo(Other /*arg*/) {} // working
};
Foo::Foo(Arg /*arg*/) {} // used not to work, fixed
Foo::Foo(Arg arg){} // working
Foo::Foo(const Arg /*arg*/) {} // working
Foo::Foo(int arg) {} // working
Change-Id: I741e4ba62672ddc99a837fdcdc27996fba5ae6c7
Reviewed-by: hjk <qthjk@ovi.com>
#!/usr/bin/env ruby
Dir.glob('**/*.cpp') { |file|
# skip ast (excluding paste, astpath, and canv'ast'imer)
next if file =~ /ast[^eip]|keywords\.|qualifiers|preprocessor|names.cpp/i
s = File.read(file)
next if s.include?('qlalr')
orig = s.dup
s.gsub!(/\n *if [^\n]*{\n[^\n]*\n\s+}(\s+else if [^\n]* {\n[^\n]*\n\s+})*(\s+else {\n[^\n]*\n\s+})?\n/m) { |m|
res = $&
if res =~ /^\s*(\/\/|[A-Z_]{3,})/ # C++ comment or macro (Q_UNUSED, SDEBUG), do not touch braces
res
else
res.gsub!('} else', 'else')
res.gsub!(/\n +} *\n/m, "\n")
res.gsub(/ *{$/, '')
end
}
s.gsub!(/ *$/, '')
File.open(file, 'wb').write(s) if s != orig
}
Change-Id: I3b30ee60df0986f66c02132c65fc38a3fbb6bbdc
Reviewed-by: hjk <qthjk@ovi.com>
Empty lists would have a single null entry, instead of being null
entirely.
Change-Id: I5114cf1e0ecdbcde5631511ad03a8d559668ea80
Reviewed-by: hjk <qthjk@ovi.com>
This time in the 'new' expression. Changed it to make
new C(1, abc...) and new C{1, abc}
work.
Change-Id: I7232798fd083b653ee04ef9ede386d6536133e16
Reviewed-by: hjk <qthjk@ovi.com>
This will fix a couple more places where brace-init-lists and
pack expansions are allowed but are not currently accepted by
the parser. For example:
foo(abc...);
now parses correctly.
Change-Id: I93710cef35154fea8437329f3174e4a2d56637b8
Reviewed-by: hjk <qthjk@ovi.com>
