This commit fundamentally changes the way symbol emission in
LDC works: Previously, whenever a declaration was used in some
way, the compiler would check whether it actually needs to be
defined in the currently processed module, based only on the
symbol itself. This lack of contextual information proved to
be a major problem in correctly handling emission of templates
(see e.g. #454).
Now, the DtoResolve…() family of functions and similar only
ever declare the symbols, and definition is handled by doing
a single pass over Module::members for the root module. This
is the same strategy that DMD uses as well, which should
also reduce the maintainance burden down the road (which is
important as during the last few releases, there was pretty
much always a symbol emission related problem slowing us
down).
Our old approach might have been a bit better tuned w.r.t.
avoiding emission of unneeded template instances, but 2.064
will bring improvements here (DMD: FuncDeclaration::toObjFile).
Barring such issues, the change shoud also marginally improve
compile times because of declarations no longer being emitted
when they are not needed.
In the future, we should also consider refactoring the code
so that it no longer directly accesses Dsymbol::ir but uses
wrapper functions that ensure that the appropriate
DtoResolve…() function has been called.
GitHub: Fixes#454.
1. Main include corresponding to .cpp file, if any.
2. DMD and LDC includes.
3. LLVM includes.
4. System includes.
Also updated a few include guards to match the default format.
This reverts commit c4adbedcc, which would have fixed the
problem at its roots, but caused strange template function
attribute inference failures in D-YAML, presumably due to
the different order of semantic3 execution on the templates.
The previous solution was problematic because the change was language-
visble via the isUnsigned trait. Now, pointers are simply explicitly
checked for in the relevant places. (Note that there might be cases in
the diff where a direct isunsigned() call would have been appropriate –
I only checked for instances where the type clearly cannot be a pointer,
but chose to go the safe route in replicating existing behavior
otherwise).
Fixes DMD testcase 'traits'.
This simplifies the code in module.cpp a bit. But it is also the base to implement a pragma to place an arbitrary function in llvm.global_ctors and llvm.global_dtors.
Previously, we just had a hack to make ref foreach statements work.
This commit enables them to work in other cases as well, like the
implicit __result variable for functions with out-contracts (which
is such a magic ref variable for ref-returning functions).
Fixes DMD testcase 'testcontracts'.
On OS X, there is an actual significance to the distinction, which before lead e.g. to exception throwing in the below example being broken:
---
import core.exception;
void main() {
asm {
jmp Lfoo;
Lfoo:
;
}
throw cast(OutOfMemoryError)cast(void*)OutOfMemoryError.classinfo.init;
assert(0);
}
---
Before, _d_arraycatT was used to concatenate multiple arrays. That caused an issue when postblit
was called on a struct multiple times. The next code asserted due to the issue:
void main()
{
static struct S
{
int x;
int pad;
this(this)
{
++x;
}
}
auto sarr = new S[1];
auto sarr2 = sarr ~ sarr ~ sarr;
assert(sarr2[0].x == 1);
assert(sarr2[1].x == 1);
assert(sarr2[2].x == 1);
assert(sarr[0].x == 0);
}
Removed use of dyn_cast, llvm no compiles
without exceptions and rtti by
default. We do need exceptions for the libconfig stuff, but rtti isn't
necessary (anymore).
Debug info needs to be rewritten, as in LLVM 2.7 the format has
completely changed. To have something to look at while rewriting, the
old code has been wrapped inside #ifndef DISABLE_DEBUG_INFO , this means
that you have to define this to compile at the moment.
Updated tango 0.99.9 patch to include updated EH runtime code, which is
needed for LLVM 2.7 as well.
generating a constant to fill the entire struct with a single `store`.
This is much more efficient at compile time (fixing #320) and vastly reduces
the size of the emitted code. Since LLVM no longer needs to keep the data for
all fields in "registers" until the store happens, it should also be more
efficient at run time in cases where the fields aren't assigned with constants.
There's also some code clean-up by removing duplicated logic.
