Don Hinton [Fri, 3 May 2019 17:47:29 +0000 (17:47 +0000)]
[CommandLine] Change help output to prefix long options with `--` instead of `-`. NFC . Part 3 of 5
Summary:
By default, `parseCommandLineOptions()` will accept either a
`-` or `--` prefix for long options -- options with names longer than
a single character.
While this change does not affect behavior, it will be helpful with a
subsequent change that requires long options use the `--` prefix.
David Blaikie [Fri, 3 May 2019 00:44:50 +0000 (00:44 +0000)]
SemaOverload: Complete candidates before emitting the error, to ensure diagnostics emitted (or suppressed) during completion don't interfere with the overload notes
Because diagnostics and their notes are not connected at the API level,
if the error message for an overload is emitted, then the overload
candidates are completed - if a diagnostic is emitted during that work,
the notes related to overload candidates would be attached to the latter
diagnostic, not the original error. Sort of worse, if the latter
diagnostic was disabled, the notes are disabled.
Richard Smith [Thu, 2 May 2019 23:21:28 +0000 (23:21 +0000)]
Fix -Wunsequenced false-positives in code controlled by a branch on
__builtin_constant_p.
If the operand of __builtin_constant_p is not constant and has
side-effects, then code controlled by a branch on it is unreachable and
we should not emit runtime behavior warnings in such code.
Leonard Chan [Thu, 2 May 2019 20:38:14 +0000 (20:38 +0000)]
[Attribute/Diagnostics] Print macro if definition is an attribute declaration
If an address_space attribute is defined in a macro, print the macro instead
when diagnosing a warning or error for incompatible pointers with different
address_spaces.
We allow this for all attributes (not just address_space), and for multiple
attributes declared in the same macro.
Nico Weber [Thu, 2 May 2019 19:47:05 +0000 (19:47 +0000)]
Another attempt to fix "could not find clang-check" lit warning in analyzer-less builds
r359717 added clang-check as a dep of check-clang unconditionally
because I had missed lit.local.cfg in test/Tooling.
Instead, only add clang-check to the tools if the analyzer is enabled,
since the build target only exists then, and since all tests using
clang-check are skipped when the analyzer is disabled.
Before this change, when emitting debug info for the
`is_same_v<int, int>` global variable, clang would crash because it
would try to use the template parameter list from the partial
specialization to give parameter names to template arguments. This
doesn't work in general, since a partial specialization can have fewer
arguments than the primary template. Therefore, always use the primary
template. Hypothetically we could try to use the parameter names from
the partial specialization when possible, but it's not clear this really
helps debugging in practice.
When the expression used to initialise 'this' has a pointer type,
check the address space of the pointee type instead of the pointer
type to decide whether an address space cast is required.
It is the pointee type that carries the address space qualifier.
Adapt -fsanitize=function to SANITIZER_NON_UNIQUE_TYPEINFO
This follows up after b7692bc3e9ad2691fc07261904b88fb15f30696b "[UBSan] Fix
isDerivedFromAtOffset on iOS ARM64" fixed the RTTI comparison in
isDerivedFromAtOffset on just one platform and then a25a2c7c9a7e1e328a5bd8274d2d86b1fadc4692 "Always compare C++ typeinfo (based on
libstdc++ implementation)" extended that fix to more platforms.
But there is another RTTI comparison for -fsanitize=function generated in
clang's CodeGenFunction::EmitCall as just a pointer comparison. For
SANITIZER_NON_UNIQUE_TYPEINFO platforms this needs to be extended to also do
string comparison. For that, __ubsan_handle_function_type_mismatch[_abort]
takes the two std::type_info pointers as additional parameters now, checks them
internally for potential equivalence, and returns without reporting failure if
they turn out to be equivalent after all. (NORETURN needed to be dropped from
the _abort variant for that.) Also these functions depend on ABI-specific RTTI
now, so needed to be moved from plain UBSAN_SOURCES (ubsan_handlers.h/cc) to
UBSAN_CXXABI_SOURCES (ubsan_handlers_cxx.h/cc), but as -fsanitize=function is
only supported in C++ mode that's not a problem.
Richard Smith [Thu, 2 May 2019 00:49:14 +0000 (00:49 +0000)]
Replace ad-hoc tracking of pattern for an instantiated class-scope
explicit function specialization with the MemberSpecializationInfo used
everywhere else.
Not NFC: the ad-hoc pattern tracking was not being serialized /
deserialized properly. That's fixed here.
Richard Smith [Thu, 2 May 2019 00:49:05 +0000 (00:49 +0000)]
Diagnose non-dependent qualified friend function template declarations
that don't match any existing declaration. Don't get confused and treat
such declarations as template *specializations*.
Tom Tan [Thu, 2 May 2019 00:38:14 +0000 (00:38 +0000)]
[COFF, ARM64] Align global symbol by size for ARM64 MSVC ABI
According to alignment section in below ARM64 ABI document, MSVC could increase
alignment of global data based on its total size. Clang doesn't do this. Compile
the same symbol into different alignments by Clang and MSVC could cause link
error because some instruction encodings, like 64-bit LDR/STR with immediate,
require the target to be 8 bytes aligned, and linker could choose code stream
with such LDR/STR instruction from MSVC and 4 bytes aligned data from Clang into
final image, which actually cannot be linked together
(see https://bugs.llvm.org/show_bug.cgi?id=41506 for more details).
Richard Trieu [Wed, 1 May 2019 23:33:49 +0000 (23:33 +0000)]
Consume unexpected "template" keywords after "using"
The parser was dealing with unexpected "template" keywords after "using"
keywords too late and putting the parser into the wrong state, which could
lead to a crash down the line. This change allows the parser to consume the
bad "template" keywords earlier, and continue parsing as if "template" was
never there to begin with for better error recovery.
Kristof Umann [Wed, 1 May 2019 19:56:47 +0000 (19:56 +0000)]
[analyzer] Don't display implementation checkers under -analyzer-checker-help, but do under the new flag -analyzer-checker-help-hidden
During my work on analyzer dependencies, I created a great amount of new
checkers that emitted no diagnostics at all, and were purely modeling some
function or another.
However, the user shouldn't really disable/enable these by hand, hence this
patch, which hides these by default. I intentionally chose not to hide alpha
checkers, because they have a scary enough name, in my opinion, to cause no
surprise when they emit false positives or cause crashes.
The patch introduces the Hidden bit into the TableGen files (you may remember
it before I removed it in D53995), and checkers that are either marked as
hidden, or are in a package that is marked hidden won't be displayed under
-analyzer-checker-help. -analyzer-checker-help-hidden, a new flag meant for
developers only, displays the full list.
Nico Weber [Wed, 1 May 2019 19:34:00 +0000 (19:34 +0000)]
Make check-clang depend on the clang-check binary always
check-clang (the target that runs all clang tests) used to
only depend on clang-check (a binary like clang-tidy,
clang-refactor, etc) if the static analyzer is enabled.
However, several lit tests call clang-check unconditionally,
so always depend on it.
Fixes a "could not find clang-check" lit warning in clean builds with
the static analyzer disabled.
Also sort the deps in the CMake file and put just one dep on each line.
[Parser] Avoid correcting delayed typos in array subscript multiple times.
We correct some typos in `ActOnArraySubscriptExpr` and
`ActOnOMPArraySectionExpr`, so when their result is `ExprError`, we can
end up correcting delayed typos in the same expressions again. In
general it is OK but when `NumTypos` is incorrect, we can hit the
assertion
> Assertion failed: (Entry != DelayedTypos.end() && "Failed to get the state for a TypoExpr!"), function getTypoExprState, file clang/lib/Sema/SemaLookup.cpp, line 5219.
Fix by replacing some subscript `ExprResult` with typo-corrected expressions
instead of keeping the original expressions. Thus if original expressions
contained `TypoExpr`, we'll use corrected expressions instead of trying to
correct them again.
Petr Hosek [Wed, 1 May 2019 16:52:45 +0000 (16:52 +0000)]
[Driver] Explicitly request platform rtlib in the Driver pic test
This test checks whether crtbegin.o and crtend.o appear on the link
line, but names of these files may be affected by the choice of the
rtlib, specifically when compiler-rt is used as the default rtlib
the names will be clang_rt.crtbegin.o and clang_rt.crtend.o instead
of crtbeginS.o and crtendS.o. To avoid the test failure, explicitly
request to use the platform rtlib.
Reference expected files not ending with a newline are normalized to
have said newlines. Additionally `plist-macros-with-expansion.cpp.plist`
is modified to add a line that is ignored by `%diff_plist`, but not by
the more sensitive pattern proposed by
http://lists.llvm.org/pipermail/cfe-dev/2019-April/061904.html for
`%normalize_plist`.
Hubert Tong [Wed, 1 May 2019 15:53:56 +0000 (15:53 +0000)]
[analyzer][tests] Use diff_plist, correct order of arguments for missed cases; NFC
For various files under `clang/test/Analysis`, D52036 applied
`%diff_plist` to replace `diff` invocations with certain options and
D56340 swapped the order of the arguments so that the reference file
comes first. The tests that used `tail` to filter the test output were
not modified accordingly. This patch applies the corresponding update
to those tests.
Tom Stellard [Wed, 1 May 2019 06:18:03 +0000 (06:18 +0000)]
lib/Header: Fix Visual Studio builds try #2
Summary:
This is a follow up to r355253 and a better fix than the first attempt
which was r359257.
We can't install anything from ${CMAKE_CFG_INTDIR}, because this value
is only defined at build time, but we still must make sure to copy the
headers into ${CMAKE_CFG_INTDIR}/lib/clang/$VERSION/include, because the lit
tests look for headers there. So for this fix we revert to the
old behavior of copying the headers to ${CMAKE_CFG_INTDIR}/lib/clang/$VERSION/include
during the build and then installing them from the source tree.
Dan Gohman [Tue, 30 Apr 2019 23:06:07 +0000 (23:06 +0000)]
[WebAssembly] Use the "wasm32-wasi" triple in tests
Similar to https://reviews.llvm.org/D61334, update clang tests to use the
"wasm32-wasi" triple, removing the "-musl" environment and omitting the
"-unknown" vendor.
Variable auto-init: don't initialize aggregate padding of all aggregates
Summary:
C guarantees that brace-init with fewer initializers than members in the
aggregate will initialize the rest of the aggregate as-if it were static
initialization. In turn static initialization guarantees that padding is
initialized to zero bits.
Quoth the Standard:
C17 6.7.9 Initialization ❡21
If there are fewer initializers in a brace-enclosed list than there are elements
or members of an aggregate, or fewer characters in a string literal used to
initialize an array of known size than there are elements in the array, the
remainder of the aggregate shall be initialized implicitly the same as objects
that have static storage duration.
C17 6.7.9 Initialization ❡10
If an object that has automatic storage duration is not initialized explicitly,
its value is indeterminate. If an object that has static or thread storage
duration is not initialized explicitly, then:
* if it has pointer type, it is initialized to a null pointer;
* if it has arithmetic type, it is initialized to (positive or unsigned) zero;
* if it is an aggregate, every member is initialized (recursively) according to
these rules, and any padding is initialized to zero bits;
* if it is a union, the first named member is initialized (recursively)
according to these rules, and any padding is initialized to zero bits;
Petr Hosek [Tue, 30 Apr 2019 19:35:14 +0000 (19:35 +0000)]
[Driver] Support compiler-rt crtbegin.o/crtend.o for Linux
When compiler-rt is selected as the runtime library for Linux targets
use its crtbegin.o/crtend.o implemenetation rather than platform one
if available.
[LibTooling] Change Transformer's TextGenerator to a partial function.
Summary:
Changes the signature of the TextGenerator std::function to return an Expected<std::string>
instead of std::string to allow for (non-fatal) failures. Previously, we
expected that any failures would be expressed with assertions. However, that's
unfriendly to running the code in servers or other places that don't want their
library calls to crash the program.
Correspondingly, updates Transformer's handling of failures in TextGenerators
and the signature of `ChangeConsumer`.
[analyzer][UninitializedObjectChecker] PR41611: Regard vector types as primitive
https://bugs.llvm.org/show_bug.cgi?id=41611
Similarly to D61106, the checker ran over an llvm_unreachable for vector types:
struct VectorSizeLong {
VectorSizeLong() {}
__attribute__((__vector_size__(16))) long x;
};
void __vector_size__LongTest() {
VectorSizeLong v;
}
Since, according to my short research,
"The vector_size attribute is only applicable to integral and float scalars,
although arrays, pointers, and function return values are allowed in conjunction
with this construct."
[src: https://gcc.gnu.org/onlinedocs/gcc-4.6.1/gcc/Vector-Extensions.html#Vector-Extensions]
[analyzer] Treat functions without run-time branches as "small".
Currently we always inline functions that have no branches, i.e. have exactly
three CFG blocks: ENTRY, some code, EXIT. This makes sense because when there
are no branches, it means that there's no exponential complexity introduced
by inlining such function. Such functions also don't trigger various fundamental
problems with our inlining mechanism, such as the problem of inlined
defensive checks.
Sometimes the CFG may contain more blocks, but in practice it still has
linear structure because all directions (except, at most, one) of all branches
turned out to be unreachable. When this happens, still treat the function
as "small". This is useful, in particular, for dealing with C++17 if constexpr.
Douglas Yung [Mon, 29 Apr 2019 22:57:15 +0000 (22:57 +0000)]
Make test more robust by writing stdout/stderr to different files.
Our internal build bots were failing this test randomly as the stderr
output was emitted to the file in the middle of the stdout output
line that the test was checking.
Simplify exclusion of nested classes from extern template instantiation, NFC
Summary:
This simplifies three checks for MS ABI, Win Itanium, or Win GNU to just
"is Windows".
The question remains, however, if this is really the correct thing to
do. We could, for example, only not consider inner classes to be
externally available if the outer class has a dllexport annotation.
However, I will leave that as future work.
Mike Rice [Mon, 29 Apr 2019 21:21:17 +0000 (21:21 +0000)]
When skipping code at the start of a file during PCH use, Preprocessor::Lex
is not used since it consumes all preprocessor directives until it returns
a real token. Using the specific Lexer (i.e. CurLexer->Lex) makes it
possible to stop skipping after an #include or #pragma hdrstop. Previously
the skipping code was only handling CurLexer, now all will be handled
correctly.
[ASTImporter] Add an ImportImpl method to allow customizing Import behavior.
Summary:
We are currently implementing support in LLDB that reconstructs the STL templates from
the target program in the expression evaluator. This reconstruction happens during the
import process from our debug info AST into the expression evaluation AST, which means
we need a way to intercept the ASTImporter import process.
This patch adds an protected ImportImpl method that we can overwrite in LLDB to implement
our special importing logic (which is essentially just looking into a C++ module that is attached to
the target context). Because ImportImpl has to call MapImported/AddToLookup for the decls it
creates, this patch also exposes those via a new unified method and checks that we call it when
importing decls.
[LibTooling] Fix unneeded use of unique_ptr where shared_ptr is expected.
Summary: This fixes a few places in the Stencil implementation where a unique_ptr is created at a callsite that expects shared_ptr. Since the former implicitly converts to the latter, the code compiles and runs correctly as is. But, there's no reason to involve unique_ptr -- the current code was leftover from a previous version in which unique_ptr was the expected type.
Ivan Donchevskii [Mon, 29 Apr 2019 13:44:07 +0000 (13:44 +0000)]
[libclang] Restore old clang_Cursor_isAnonymous behaviour
D54996 Changed the behaviour of clang_Cursor_isAnonymous, but there is no alternative available to get the old behaviour in some cases, which is essential for determining if a record is syntactically accessible, e.g.
struct {
int x;
int y;
} foo;
struct {
struct {
int x;
int y;
};
} bar;
void fun(struct { int x; int y; } *param);
The only 'anonymous' struct here is the one nested in bar, since there is
no way to reference the struct itself, only the fields within. Though the
anonymity applies to the instance itself, not the type.
To avoid confusion, I have added a new function called clang_Cursor_isAnonymousRecordDecl
which has the old behaviour of clang_Cursor_isAnonymous (and updated the doc
for the latter as well, which was seemingly forgotten).
Richard Smith [Sat, 27 Apr 2019 02:58:17 +0000 (02:58 +0000)]
Reinstate r359059, reverted in r359361, with a fix to properly prevent
us emitting the operand of __builtin_constant_p if it has side-effects.
Original commit message:
Fix interactions between __builtin_constant_p and constexpr to match
current trunk GCC.
GCC permits information from outside the operand of
__builtin_constant_p (but in the same constant evaluation context) to be
used within that operand; clang now does so too. A few other minor
deviations from GCC's behavior showed up in my testing and are also
fixed (matching GCC):
* Clang now supports nullptr_t as the argument type for
__builtin_constant_p
* Clang now returns true from __builtin_constant_p if called with a
null pointer
* Clang now returns true from __builtin_constant_p if called with an
integer cast to pointer type
Petr Hosek [Sat, 27 Apr 2019 00:25:13 +0000 (00:25 +0000)]
[Fuchsia] Support multilib for -fsanitize=address and -fno-exceptions
This introduces a support for multilibs to Fuchsia driver. Unlike the
existing multilibs that are used primarily for handling different
architecture variants, we use multilibs to handle different variants
of Clang runtime libraries: -fsanitize=address and -fno-exceptions
are the two we support initially. This replaces the existing support
for sanitized runtimes libraries that was only used by Fuchsia driver
and it also refactors some of the logic to allow sharing between GNU
and Fuchsia drivers.
Petr Hosek [Sat, 27 Apr 2019 00:25:11 +0000 (00:25 +0000)]
[Driver] Support priority for multilibs
When more than one multilib flag matches, try to select the best
possible match based on priority. When two different multilibs with
the same same priority match, we still throw an error matching the
existing behavior.
Alex Lorenz [Fri, 26 Apr 2019 22:40:47 +0000 (22:40 +0000)]
[driver][macOS] Link libarclite from the default toolchain when clang
is running in a toolchain outside of xcode
'libarclite' usually lives in the same toolchain as 'clang'. However, the
Swift open source toolchains for macOS distribute Clang without 'libarclite'.
In that case, to allow the linker to find 'libarclite', we point to the
'libarclite' that should be in the XcodeDefault toolchain instead. The
path to the toolchain is inferred from the SDK path if it's specified.
[AArch64] Add support for MTE intrinsics
This provides intrinsics support for Memory Tagging Extension (MTE),
which was introduced with the Armv8.5-a architecture.
These intrinsics are available when __ARM_FEATURE_MEMORY_TAGGING is defined.
Each intrinsic is described in detail in the ACLE Q1 2019 documentation:
https://developer.arm.com/docs/101028/latest
Reviewed By: Tim Nortover, David Spickett
Differential Revision: https://reviews.llvm.org/D60485
Martin Storsjo [Fri, 26 Apr 2019 19:31:51 +0000 (19:31 +0000)]
[MinGW] Always emit local typeinfo
This makes sure that code built with headers for a statically linked
libc++ also works when linking to the DLL version, when the DLL
hasn't been built with --export-all-symbols.
This matches what GCC for MinGW does for this test case.
Martin Storsjo [Fri, 26 Apr 2019 19:31:46 +0000 (19:31 +0000)]
[MinGW] Do dllexport inline methods in template instantiation
Normally, in MinGW mode, inline methods aren't dllexported.
However, in the case of a dllimported template instantiation,
the inline methods aren't instantiated locally, but referenced
from the instantiation. Therefore, those methods also need to
be dllexported, in the case of an instantiation.
GCC suffers from the same issue, reported at [1], but the issue
is still unresolved there.
Martin Storsjo [Fri, 26 Apr 2019 19:31:39 +0000 (19:31 +0000)]
[MinGW] Don't let template instantiation declarations cover nested classes
An explicit template instantiation declaration used to let
callers assume both outer and nested classes instantiations were
defined in a different translation unit.
If the instantiation is marked dllexport, only the outer class
is exported, but the caller will try to reference the instantiation
of both outer and inner classes.
This makes MinGW mode match both MSVC and Windows Itanium, by
having instantations only cover the outer class, and locally emitting
definitions of the nested classes. Windows Itanium was changed to
use this behavious in SVN r300804.
This deviates from what GCC does, but should be safe (and only
inflate the object file size a bit, but MSVC and Windows Itanium
modes do the same), and fixes cases where inner classes aren't
dllexported.
This fixes missing references in combination with dllexported/imported
template intantiations.
GCC suffers from the same issue, reported at [1], but the issue
is still unresolved there. The issue can probably be solved either
by making dllexport cover all nested classes as well, or this
way (matching MSVC).
[OPENMP]Added check for non-random access types for the dependent loop
counters.
According to the OpenMP 5.0, For any associated loop where the b or lb
expression is not loop invariant with respect to the outermost loop, the
var-outer that appears in the expression may not have a random access
iterator type.
[ASTImporter] Copy Argument Passing Restrictions setting when importing a CXXRecordDecl definition
Summary:
For a CXXRecordDecl the RecordDeclBits are stored in the DeclContext. Currently when we import the definition of a CXXRecordDecl via the ASTImporter we do not copy over this data.
This change will add support for copying the ArgPassingRestrictions from RecordDeclBits to fix an LLDB expression parsing bug where we would set it to not pass in registers.
Note, we did not copy over any other of the RecordDeclBits since we don't have tests for those. We know that copying over LoadedFieldsFromExternalStorage would be a error and that may be the case for others as well.
The companion LLDB review: https://reviews.llvm.org/D61146
Yonghong Song [Fri, 26 Apr 2019 15:35:51 +0000 (15:35 +0000)]
[BPF] do not generate predefined macro bpf
"DefineStd(Builder, "bpf", Opts)" generates the following three
macros:
bpf
__bpf
__bpf__
and the macro "bpf" is due to the fact that the target language
is C which allows GNU extensions.
The name "bpf" could be easily used as variable name or type
field name. For example, in current linux kernel, there are
four places where bpf is used as a field name. If the corresponding
types are included in bpf program, the compilation error will
occur.
This patch removed predefined macro "bpf" as well as "__bpf" which
is rarely used if used at all.
Signed-off-by: Yonghong Song <yhs@fb.com>
Differential Revision: https://reviews.llvm.org/D61173
Martin Storsjo [Fri, 26 Apr 2019 08:09:51 +0000 (08:09 +0000)]
[MinGW] Fix dllexport of explicit template instantiation
Contrary to MSVC, GCC/MinGW needs to have the dllexport attribute
on the template instantiation declaration, not on the definition.
Previously clang never marked explicit template instantiations as
dllexport in MinGW mode, if the instantiation had a previous
declaration, regardless of where the attribute was placed. This
makes Clang behave like GCC in this regard, and allows using the
same attribute form for both MinGW compilers.
Adam Balogh [Fri, 26 Apr 2019 07:30:07 +0000 (07:30 +0000)]
[Analyzer] Iterator Checkers - Do an early return after handling calls
This patch is more of a fix than a real improvement: in checkPostCall()
we should return immediately after finding the right call and handling
it. This both saves unnecessary processing and double-handling calls by
mistake.
[analyzer] RetainCount: Add a suppression for "the Matching rule".
In the OSObject universe there appears to be another slightly popular contract,
apart from "create" and "get", which is "matching". It optionally consumes
a "table" parameter and if a table is passed, it fills in the table and
returns it at +0; otherwise, it creates a new table, fills it in and
returns it at +1.
For now suppress false positives by doing a conservative escape on all functions
that end with "Matching", which is the naming convention that seems to be
followed by all such methods.
[analyzer] RetainCount: Allow offsets in return values.
Because RetainCountChecker has custom "local" reasoning about escapes,
it has a separate facility to deal with tracked symbols at end of analysis
and check them for leaks regardless of whether they're dead or not.
This facility iterates over the list of tracked symbols and reports
them as leaks, but it needs to treat the return value specially.
Some custom allocators tend to return the value with an offset, storing
extra metadata at the beginning of the buffer. In this case the return value
would be a non-base region. In order to avoid false positives, we still need to
find the original symbol within the return value, otherwise it'll be unable
to match it to the item in the list of tracked symbols.
[analyzer] Fix crash when returning C++ objects from ObjC messages-to-nil.
the assertion is in fact incorrect: there is a cornercase in Objective-C++
in which a C++ object is not constructed with a constructor, but merely
zero-initialized. Namely, this happens when an Objective-C message is sent
to a nil and it is supposed to return a C++ object.
Richard Smith [Fri, 26 Apr 2019 01:51:08 +0000 (01:51 +0000)]
C++ DR2387: a variable template declared wtih (or instantiated with) a
const-qualified type is not implicitly given internal linkage. But a
variable template declared 'static' is.
This reinstates part of r359048, reverted in r359076.
Such constructs are ill-formed by [temp.explicit]p13. We make a special
exception to permit an invalid construct used by libc++ in some build
modes: its <valarray> header declares some functions with the
internal_linkage attribute and then (meaninglessly) provides explicit
instantiation declarations for them. Luckily, Clang happens to
effectively ignore the explicit instantiation declaration when
generating code in this case, and this change codifies that behavior.
This reinstates part of r359048, reverted in r359076. (The libc++ issue
triggering the rollback has been addressed.)