fix Issue 21282 mixin of AliasSeq cannot alias an expression

[WalterBright]

No description provided.

[dlang-bot]

Thanks for your pull request, @WalterBright!

Bugzilla references

Auto-close	Bugzilla	Severity	Description
✓	21282	regression	mixin of AliasSeq "cannot alias an expression"

⚠️⚠️⚠️ Warnings ⚠️⚠️⚠️

• Regression or critical bug fixes should always target the stable branch. Learn more about rebasing to stable or the D release process.

To target stable perform these two steps:

1. Rebase your branch to upstream/stable:

git rebase --onto upstream/stable upstream/master

2. Change the base branch of your PR to stable

Testing this PR locally

If you don't have a local development environment setup, you can use Digger to test this PR:

dub run digger -- build "master + dmd#11810"

[thewilsonator]

This is a regression. It should target stable.

[BorisCarvajal]

@WalterBright I was fixing this issue just now and your fix seems very overkill.
The mine is just a little change to aliasSemantic:

if (e)  // Try to convert Expression to Dsymbol
{
    if (auto te = e.isTupleExp())
         s = new TupleDeclaration(te.loc, ds.ident, cast(Objects*)te.exps);
    else
...

Currently you can alias a tuple but only if it's encapsulated as a symbol (TupleDeclaration).
The previous change to mixin exposed a new code path where the tuple is presented as an expression to the alias code.
My fix is simple, just create a TupleDeclaration from the TupleExpression.

[WalterBright]

@BorisCarvajal where is the mixin be expanded? Anyhow, create a PR for your version, and we'll compare.

[WalterBright]

Buildkite has ANOTHER Heisenbug:

[boostrap] Downloading compiler http://downloads.dlang.org/releases/2.x/2.088.0/dmd.2.088.0.linux.tar.xz
--
  | curl: (56) Recv failure: Connection reset by peer
  | xz: (stdin): Unexpected end of input
  | tar: Unexpected EOF in archive
  | tar: Unexpected EOF in archive
  | tar: Error is not recoverable: exiting now
  | ERROR: bootstrapping failed.
  | 🚨 Error: The command exited with status 1

https://issues.dlang.org/show_bug.cgi?id=21284

Is it possible we can get some Test Suite action on these problems? It's really not good enough to have to keep restarting the whole shebang.

https://issues.dlang.org/buglist.cgi?bug_status=NEW&bug_status=ASSIGNED&bug_status=REOPENED&keywords=TestSuite%2C%20&keywords_type=allwords&list_id=233345&query_format=advanced

[BorisCarvajal]

where is the mixin be expanded?

I think in typesem.d->resolve()->visitMixin()

Gonna push it soon.

[thewilsonator]

That looks like a network error, good luck eliminating them entirely.

[WalterBright]

That looks like a network error, good luck eliminating them entirely.

A number of them have been successfully fixed with the method: for network errors, sleep for a few seconds, then try it again.

The problem with not fixing these is the number of tests we have that are dependent on the network is getting so large it becomes a virtual certainty that it will fail somewhere and we can't get a clean test run.

[BorisCarvajal]

Oops, the bug is not tested.

[WalterBright]

ak, you're right

[BorisCarvajal]

found it when I was stealing the test case.

[BorisCarvajal]

my pull #11814

[WalterBright]

Closing in favor of #11814