EvalFunctionCall.fImplicitThis is sometimes redundant in that the
owner evaluation is already stored by one of the arguments. In
such cases, storing the owner separately in fImplicitThis can lead
to exponential complexity in chained method calls.
We resolve the duplication by computing the implicit this from the
function name evaluation instead of storing it where possible.
This was already implemented for cases where the function name
evaluation is an EvalMemberAccess in commit 659ff8c4a7. This
commit extends the approach to cases where the function name
evaluation is an EvalID.
Change-Id: Ic71e81b4692c51ffb8e15b3da9fc2dff1a554f05
When a subsequent regular (non-friend) declaration of such a class
is indexed, the index binding needs to be marked as being fully
visible to name lookup.
Change-Id: I1a625f93eda1af257a9af50b5c4f115fc9bf6526
The first patch for bug 527697 made us not instantiate such an
argument, because determinePackSize() would return PACK_SIZE_DEFER.
The motivation for that fix was to avoid sizeof...(T) prematurely
instantiating to a concrete value in cases where T was mapped to
a pack expansion.
This patch reverts the change to determinePackSize() and applies a
different fix for the sizeof...(T), specific to EvalUnaryTypeId.
Change-Id: Idc231aeecb5d50e93dda364c6d2deb08057cc8b6
- default environment using EnvironmentReader in
CommandLauncherWrapper class in CommandLauncherManager
if no environment set for wrapper to match behaviour
of CommandLauncher
Change-Id: I71ff37b3d8f2049dc705e97bf28f939ac78bd390
This avoids expontential complexity when type template arguments inside
an ambiguous name specifier themselves contain ambiguous name specifiers.
The patch also enhances TemplateIdStrategy to allow marking and backing
up to a branch point, and uses this ability in templateArgument().
Change-Id: Ia03e9cd0bc026b02b85edc05ed327cce883d6a59
The special binding type CPPLambdaExpressionParameter is removed.
Instead, a lambda expression parameters's are represented as
regular CPPParameters owned by the closure type's generated
function call operator.
Change-Id: I4afeac90c2595a1f84dfa59f057d0494b64d079c
The pseudo-destructor is represented as a CPPImplicitFunction, computed
lazily and stored by CPPBasicType.
This commit also adds support for alias templates to
CPPTemplates.getTemplateName().
Change-Id: I6774556b2493cb68d32c3007d6ce48c7805595f4
Previously, we would only try the first base class whose primary
template matched that of the parameter type.
Change-Id: I0511e6a1ba1c7197887ff23bc37b70a2a820eb87
- Add volatile variant for __atomic_clear and __atomic_test_and_set
- Add missing parameters for __builtin_fmax and related
Change-Id: Ib31d1a410b8749f5672d70d7d02ca1dff962c38b
Signed-off-by: Martin Oberhuber <mober.at+eclipse@gmail.com>
The evaluated type of 'decltype(auto)' in combination with const and/or
volatile will be a ProblemType since this is not valid code. The patch
also contains a checker to give the user a visual feedback.
Note: A proposed quick-fix has been removed after a short discussion.
Change-Id: I8760ed0ac28e28529ab30516accac9c0413c87d9
Signed-off-by: Hansruedi Patzen <hansruedi.patzen@hsr.ch>
When deleting projects then creating them with the same name
the scanner info didn't get cleared up and was reused. We
were deleting these files from the wrong path.
Change-Id: Ieda174b7cfc9690ab51be65e674bda321cb3bc10
We were missing the dialog to ask if you wanted to continue
after build errors in a project. We were missing the method
that provided the list of projects to do that check.
Also while testing that noticed the handling of the build
and clean command UI was broken with make projects.
Change-Id: I698e151672a114bb22c815f49d362b6413b9c315
Makefile projects depend on build output parsing to pick up
compiler options to use while indexing. Sometimes, that just
doesn't work. As a back up we've always allowed users to manually
add them to the project. Long ago when we had pure make projects
we had UI for this. This change hooks that old UI up. It was a
time before build configs so they get applied to them all but it
serves it's role as a backup when users run into problems.
Unmarked everything I use here from deprecated until we decide
how to replace these with Core build config aware settings.
Change-Id: I6fc4d48e5315df3c60c7591b026ae0a53a1f505e
Also set BUILD_MODE to the launch mode so the makefile can do
different things based on the mode. Change template to add -g
option for debug.
Also fix a deadlock I detected due to side affect build folder
creation.
Change-Id: Ic7b13ba3238e6ef201cccb1b2bfc8dcc6956ea3a
The previous implementation deviated from the C++ standard by checking
that the types of the return expressions are the same, rather than the
return types after deduction against the placeholder type.
There was also a bug in the return type deduction code for lambdas,
where for a lambda without an explicit placeholder in the trailing-
return-type, the deduction process wouldn't be performed.
Change-Id: I2f0b9f1c7778aef60e4cd7ada9386b99be52669a
The computation had a bug where the array decayed to a pointer, and we
tried to use the pointer's value as a composite value, instead of the
underlying array's value.
Change-Id: I9510d28e04deb0b8ef835e2857f8b513d11d1d72
This works around the fact that the optimization introduced in bug 316704
inteferes with the mechanism for offering completions for both alternatives
in an ambiguous context.
Change-Id: Ibe14c1b4f2f9c9b3394d4635c87424a25fbd7a53
Using ESP-IDF (the ESP32 FreeRTOS SDK) on MSYS2 as a test bed.
Making sure the indexer picks up as much as it can. Removed -j
since that messes up build output parsing. Add UI so you can put
it back if you want.
Change-Id: I767c739dce1412c75fb56d0bb1efceb913883a5f
Often Makefile projects are intended to be built from the project
root directory, such as ESP32 IDF projects. This adds a setting to
control that.
Change-Id: I151ebf60a7609461d24e192a9d6418fccb737055
- the CommandLauncherManager.CommandLauncherWrapper class is
designed to figure out what launcher to use after being
created and a project is set, but it is not built for reuse
- change the CommandLauncherWrapper class to reset the launcher
back to null when the project gets set so that any reuse
will be sure to get the correct internal launcher after
resetting the project
Change-Id: I2c0758ca53f7164d4bbd4a6f738acd7c7c042370
Reuses the old makeNature. Reuses the StandardBuildConfiguration.
Generates a pretty simple project for now. Also handles the case
where you don't want to generate anything, just create an empty or
on an existing source tree.
Change-Id: I2f3cddc85d55792a2c537e37d4bc236a3073d930
InstantiationContext.setExpandPack() and related methods were introduced
in bug 486971 to ensure that when instantiating a type list that
contains a pack expansion, with a parameter map that maps the template
parameter pack that appears in the expansion to another parameter pack
(which can happen when e.g. instantiating an alias template with
dependent arguments), the pack is expanded in the correct place.
However, bug 486971 only added use of this machinery to CPPTemplates.
instantiateArguments(). We can also instantiate a type list in
instantiateTypes() (used e.g. when instantiating the parameter types
of a function type), so the machinery needs to be used there as well.
Change-Id: Iabb458e8e3166c15ed922656fc0729a4a8cf8bbf
The new Autotools core build executes autoreconf which is a perl
script. We've been assuming up until now they were always
Windows exes or bats. If there's no proper extension, try
finding sh and passing the command to it.
Change-Id: I71ba66d7658db0bdc45608abc356b5efe80669af
