Now `mir_const_qualif` must be called for `static`s and `const fn`s as
well as `const`s since it is responsible for const-checking. We return
the qualifs in the return place for everything, even though they will
only be used for `const`s.
Unlike the original pass, we check *every* non-cleanup basic block
instead of stopping at `SwitchInt`. We use the `is_cfg_cyclic` function
to check for loops unlike the original checker which could not differentiate
between true cycles and basic blocks with more than two predecessors.
The last three functions are all copied verbatim from `qualify_consts`.
libpanic_unwind for Miri: make sure we have the SEH lang items when needed
r? @oli-obk @alexcrichton This is required to fix the Miri toolstate. Turns out rustc complains when doing codegen for MSVC and these lang items do not exist. For now `cfg(miri)` needs to still be able to codegen (we [plan to change that](https://github.com/rust-lang/miri/pull/1048#issuecomment-554108470) but that's a larger project requiring improvements to xargo and maybe also cargo; that should not block fixing the toolstate). Yes, this is a hack, but it is inside `cfg(miri)` so I hope this is okay.
Cc @Aaron1011
Move the JSON error emitter to librustc_errors
This is done both as a cleanup (it makes little sense for this emitter to be in libsyntax), but also as part of broader work to decouple Session from librustc itself.
Along the way, this also moves SourceMap to syntax_pos, which is also nice for the above reasons, as well as allowing dropping the SourceMapper trait from code. This had the unfortunate side-effect of moving `FatalError` to rustc_data_structures (it's needed in syntax_pos, due to SourceMap, but putting it there feels somehow worse).
Add --force-run-in-process unstable option to libtest
When running tests with `-Zpanic_abort_tests`, it's sometimes desirable to fall back to the old behavior of only running tests in-process. This comes in handy if the system process launcher is unavailable, or the test code somehow expects all tests to be run in the same process.
For example, in Fuchsia we have unit tests that actually test the process launcher itself, in which case we can't use the process launcher to run the tests :).
r? @alexcrichton
cc @cramertj,@petrhosek
Add more tests for fixed ICEs
Closes#36122 (fixed in 1.20.0)
Closes#58094 (fixed in #66054)
Also, fix mistaken test case, from #30904 to #30906 (cc @eddyb)
r? @Centril
That condition was leftover from a refactor, and was probably not
intended. In fact it can't trigger: it would require a ConstantValue of
an integral type for which `try_eval_bits` fails. But since we only
apply `subtract_ctors` to the output of `all_ctors`, this won't happen.
