Update frontend to dmd/stable

[ibuclaw]

Only got as far as linking, no testing runtime problems yet...

[ibuclaw]

Compiler seems fine now when building the parts of druntime and phobos it can. No compile-time ICE or failures when building druntime/phobos. Only reported missing symbols in object.d

[Geod24]

That's 2.081 right ? 👍

[ibuclaw]

@Geod24 unless a release happened in the last seven hours since I looked, then it's still 2.081-beta.

[ibuclaw]

The first patch should be split up between frontend and glue changes. Some parts can be cherry picked into the C++ branch.

[ibuclaw]

There's a number of ICEs introduced by the new frontend. Surprisingly a couple look suspiciously like memory corruption. Possibly coming from dinterpret.d, where in one example there's a StructLiteralExp with pointers to what was then the stack.

# of expected passes		30806
# of unexpected failures	296
# of unresolved testcases	2924

[ibuclaw]

This 3 dozen commit mess is the reason why I don't do "release" merges anymore. Still many bugs to fix, still memory corruption bug in interpret3.d, many implementation regressions pushed upstream though.

# of expected passes		31005
# of unexpected failures	44
# of unresolved testcases	2868

[ibuclaw]

Thousands of unresolved tests were due to a link error have seen previously (there's an open upstream bug report on it).

Reverted, and now we're down to something more manageable.

# of expected passes		33729
# of unexpected failures	76
# of unresolved testcases	112

[ibuclaw]

Looks like just around 4 things left to look at.

• ICE in testsuite caused by memory corruption.
• ICE in building unittests - bad dwarf codegen by the looks.
• Check for new runtime library functions that should be called, add missing tests.
• Resync with upstream, as its been another week since last.

[ibuclaw]

ICE in unittests is reproducible on master:

void test()
{
  interface I {
    final void foo() { }
  }
}

Happens because dwarf2out doesn't handle DW interface types in prune_unused_types_walk_local_classes(), added case to the gcc patches, however alternatively we could just not mark D interface types with RECORD_IS_INTERFACE in the first place, so its just treated as a regular class according to debug info.

[ibuclaw]

Looks like just the memory corruption to address.

[belka-ew]

I think, I found a regression:

module dub.generators.generator;

struct Package
{
    @property string name() const
    {
        return null;
    }
}

void ProjectGenerator()
{
    string[string] targets;

    void collectDependencies(ref string)
    {
    }

    collectDependencies(targets[Package.init.name]);
}

gives:

source/dub/generators/generator.d: In function ‘ProjectGenerator’:
source/dub/generators/generator.d:19:24: internal compiler error: in add_expr, at tree.c:7454
collectDependencies(targets[Package.init.name]);
^
0x6f7308 ???
../sysdeps/x86_64/start.S:118

[ibuclaw]

@belka-ew - I can reproduce on stable branch, so was not introduced by this change.

Should raise a bug report though.

[belka-ew]

ok. everything else looks fine

[ibuclaw]

OK, hitting merge, have plenty of time to fix bugs and iron out other parts. At least with this in we can do weekly merge updates with dmd. :-)

Though having a quick look at the size/frequency of master->stable merges, will probably want to switch to being in sync with master when the next merge happens.

dlang/dmd@4473071 88 days ago
dlang/dmd@b591bdb 29 days ago

As the idea was to catch problems introduced by dmd early in the development cycle.