make get for options use lent T

[cooldome]

fixes #14420

[timotheecour]

I'm confused; what about #14417 (comment) ? at very least this needs a test at RT and CT
also the title of #14420 is The VM doesn't support lent annotation, so unless you're fixing that (which doesn't look like it from PR content), I'd remove fixes #14420 from PR msg

[cooldome]

@mratsim tried adding lent T to get() function of Option[T] and received error Error: limited VM support for 'addr' and thought that VM has no support for lent T. It turned out that VM has full support of lent T/var T and the issue is caused not by lent T but by temporary variable introduced by vmgen to handle stmtListExpr. One miinor change to get() was needed to avoid stmtListExpr.

Option[T] is already extensively covered by tests.

[mratsim]

Is the temporary introduced by the VM needed?

Otherwise maybe we should add "Try returning via the implicit result" to the error message.

unsafeGet should also be updated in this PR.

[timotheecour]

• ok thanks for explanation; I've added a reduced example for this in VM and js backend require result for lent or var annotations to work #14420 (comment)

We should fix #14420 (comment) before closing that issue, so I still don't think your PR should close that issue as it remains an incompatibility between RT and VM

• @mratsim can you re-evaluate VM and js backend require result for lent or var annotations to work #14420 (comment) after the changes in this PR?

As a potential impact, I tried to use options in my RayTracer, but they were 5x slower (not just a couple dozen percents like the iterator issue in #14421) than passing mutable parameter + bool
Well nobody claimed Option[T] is a lightweight abstraction.

• Option[T] is already extensively covered by tests.

there's 0 VM test

[arnetheduck]

this doesn't fix the issue though, just works around it for Option specifically - ie result = self.val is just a hack, the same construct exists in lots and lots of other code.