[ltsmaster] Backport PR #2315. #2334
Conversation
|
@redstar As you're busy with lts master, is this PR still valuable? Or should we just drop debuginfo support in lts (because we only use it as bootstrapper)? I reckon debuginfo may help with fixing bugs but... |
|
Yes, I consider this PR still valueable. One of my next tasks. I like to have ltsmaster in a good, clean state. |
|
@JohanEngelen I do not get the test failure ( |
The change is to fix debuginfo build errors with Phobos and LLVM5.0. This PR is mostly copy-pasting from master and fixing compile errors, so not much checking whether things make sense or not for LTS.
JohanEngelen
force-pushed
the
fixdi50
branch
from
2456bcf
to
8993eb7
Compare
|
@redstar done |
|
I now managed to reproduce the error on one system. |
| IF_LOG { | ||
| Logger::cout() << "Addr: " << *val << '\n'; | ||
| Logger::cout() << "of type: " << *val->getType() << '\n'; | ||
| } | ||
| if (byref || (vd->isParameter() && getIrParameter(vd)->arg && | ||
| getIrParameter(vd)->arg->byref)) { | ||
| const bool isRefOrOut = vd->isRef() || vd->isOut(); |
There was a problem hiding this comment.
Evaluation of isRefOrOut and vd->isParameter... gives different results and as consequence a missing load, leading to the failure in the test suite. I think a simple solution is to add
const bool isRefParameter = vd->isParameter() && getIrParameter(vd)->arg && getIrParameter(vd)->arg->byref;
and change the condition below to
} else if (byref || isRefOrOut || isRefParameter) {
This makes sure that no load is missed.
There was a problem hiding this comment.
@redstar you can push to this branch (afaik), can you add your fix?
|
I merged your commit. |
The change is to fix debuginfo build errors with Phobos and LLVM5.0. This PR is mostly copy-pasting from master and fixing compile errors, so not much checking whether things make sense or not for LTS.