CircleCI: Bump jobs to Ubuntu 24

[kinke]

No description provided.

[kinke]

FYI @liushuyu. Looking pretty good, just some sanitizer problems with the distro sanitizer libs, the buggy codegen/xtensa.d that you've fixed in the LLVM 19 PR (IIRC), and otherwise just a single gdb-related failure with gdb v15.0, for a missed global again (imported from another module, referenced by identifier only):

ldc/tests/debuginfo/print_gdb.d

Line 100 in 40716be

// GDB: p b_Glob

[liushuyu]

just some sanitizer problems with the distro sanitizer libs

This is due to Docker blocking the ASLR disable request (because it's a security risk). See moby/moby#43011. The temporary workaround is to disable the container security features (or enable privileged).

otherwise just a single gdb-related failure with gdb v15.0, for a missed global again (imported from another module, referenced by identifier only):

(gdb) p inputs.import_b.b_Glob
$2 = 99.8799973

This works. I don't know if this should be the correct behaviour, though.

[liushuyu]

@kinke

Work around possible remaining gdb v15 regression in tests/debuginfo/print_gdb.d
1dd76cc

I think I will still open a bug report for GDB, as LLDB (which does not have D support) can resolve this symbol without the namespace:

* thread #1, name = 'print_gdb', stop reason = breakpoint 1.1
    frame #0: 0x00005555555552f2 print_gdb`D main at print_gdb.d:103:5
   100  // GDB: p b_Glob
   101  // CHECK: = 99.8
   102
-> 103      cC clsC = new cC;
   104      clsC.classMemberFunc(2);
   105      cC.classStaticFunc(4);
   106      clsC.mixedVal++;
(lldb) p b_Glob
error: print_gdb [0x000000000000002f]: unhandled type tag DW_TAG_module (DW_TAG_module), please file a bug and attach the file at the start of this error message
(float) 99.8799972
(lldb)

(Although it might look like LLDB did not correctly handle module declaration)

[liushuyu]

Congratulations on making the scripts run on someone else's computer (aka CI)!

I think you can just cherry-pick 8387651 to make the remainder of the CI tests pass.

[kinke]

I think I will still open a bug report for GDB

Thanks, that'd be awesome. As it worked with gdb 9 too, I'm pretty sure it should work without fully qualified name.

Congratulations on making the scripts run on someone else's computer (aka CI)!

Heh thanks, your pointers were very helpful.

I think you can just cherry-pick 8387651 to make the remainder of the CI tests pass.

[I don't see the Circle failures in your LLVM-19 PR as a blocker for merging, nor the xtensa failure here as a blocker for this PR, Both PRs are ready and should land very soon, just giving Nicholas some time to reply wrt. the SPIR-V target layout.] I didn't want to cherry-pick the xtensa fix with other unrelated changes, nor apply your fix manually and potentially cause merge conflicts then for you.

[kinke]

For posterity: container and VM performances seem more or less identical, and the credit consumption is identical too. The only real downside of the VM is the startup time, something like 30-60 seconds.

[liushuyu]

Thanks, that'd be awesome. As it worked with gdb 9 too, I'm pretty sure it should work without fully qualified name.

@kinke I have opened a bug report on the upstream and a suggested fix: https://sourceware.org/bugzilla/show_bug.cgi?id=32398. I can add your email address (as seen from GitHub) to the cc list if you want (so that you can test the fix upstream provided); I know some companies have specific rules when using the company email, so I won't do it without asking.

[kinke]

Oh wow, so you did basically everything already! :) Thanks! - I fully trust your detailed report to be enough for the gdb team to come up with a regression test case; I don't build gcc myself or use experimental toolchains, so please leave me out, I won't be able to early-test. But please link your report in #4389, which we'll be able to close once the gdb fix has landed and we can confirm it via CI. It's not an LDC issue per se as you know, just a place where this testsuite-problem is tracked, and remains open, whereas this PR is merged, closed and forgotten. :)

[liushuyu]

Oh wow, so you did basically everything already! :) Thanks! - I fully trust your detailed report to be enough for the gdb team to come up with a regression test case; I don't build gcc myself or use experimental toolchains, so please leave me out, I won't be able to early-test. But please link your report in #4389, which we'll be able to close once the gdb fix has landed and we can confirm it via CI. It's not an LDC issue per se as you know, just a place where this testsuite-problem is tracked, and remains open, whereas this PR is merged, closed and forgotten. :)

Thanks! Will do.