Rainer Schuetze

> We have a bison implementation of the abi spec somewhere iirc, I'll see if I can find it. There are a couple of versions here: https://gist.github.com/rainers/6cdf73b48837defb9f88 showing different attempts...

> Which adds only several more conflicts to your existing 53. I guess you are using one of the website-grammars. These don't work reasonably well with LALR(1). IIRC backref6.bison is...

> I did use one initially, but took it out to make it shorter. Does it have to be LALR(1)? I don't see why. It doesn't have to, but it...

> IIRC backref6.bison is the latest version that works without conflicts Well, it does so only because the mangling of extern(Pascal) and extern(Objective-C) had been replaced with some proposal I...

Tests fail because options from `DFLAGS` are appended to the command line arguments and are now ignored for compilation. As you suggested, maybe it's just better to bail out if...

``` make[1]: *** [posix.mak:312: generated/freebsd/debug/64/unittest/libdruntime-ut.so] Segmentation fault (core dumped) ``` Isn't this clear enough? It crashes building druntime unittests.

I guess it has been suggested before, but maybe it would be better to make ``` int[int] aa = []; ``` allocate the internal data struct. It currently yields a...

`new AA` sounds good to me too. With `alias AA = int[string]` it currently errors out with `new can only create structs, dynamic arrays or class objects, not int[string]'s` and...

I think calling abort if the GC isn't available is the right approach. Using a statically allocated AssertError could be ok, too, but should probably be thread local. It needs...

> Could something along the lines of https://github.com/dlang/druntime/blob/v2.075.0-b4/src/rt/deh.d#L22 Using the initializer is just one possibilty, even a pretty bad one, because you have to modify it to store location information...