syntax: Disallow using @goto to skip finally blocks

[IanButterworth]

Add syntax errors when @goto is used to jump out of try, catch, or else blocks when a finally block is present.

Fixes #60972

[Keno]

In the unlikely event that pkgeval comes back negative, I will note that it is possible to use syntax evolution to detect the syntax version of the calling module.

[oscardssmith]

Triage likes this (especially that it parallels goto within try with finally)

julia> function f()
           try
               @goto hi
           catch
               return 1
           finally
              return 3
           end
           @label hi
           return 2
       end
ERROR: syntax: goto from a try/finally block is not permitted around REPL[6]:2
Stacktrace:
 [1] top-level scope
   @ REPL[6]:1

Triage also thinks that this should probably done via syntax evolution (potentially with a deprecation warning). We built ourselves a mechanism for evolving the language syntax so we should be using it.

[oscardssmith]

@nanosoldier runtests()

[nanosoldier]

The package evaluation job you requested has completed - possible new issues were detected.
The full report is available.

Report summary

❗ Packages that crashed

281 packages crashed on the previous version too.

✖ Packages that failed

18 packages failed only on the current version.

• Package has test failures: 4 packages
• Tests became inactive: 1 packages
• Test duration exceeded the time limit: 13 packages

1229 packages failed on the previous version too.

✔ Packages that passed tests

8 packages passed tests only on the current version.

• Other: 8 packages

5573 packages passed tests on the previous version too.

~ Packages that at least loaded

3413 packages successfully loaded on the previous version too.

➖ Packages that were skipped altogether

3 packages were skipped only on the current version.

• Package could not be installed: 3 packages

906 packages were skipped on the previous version too.

[PallHaraldsson]

I looked at all of (these) failures (probably all false alarms);

EliminateGraphs | v0.2.1 | fail | test | ▇▇▇▇▇▇▇▇▇▇▇
PushVectors | v0.2.1 | fail | test | ▇▇▇▅▇▇▇▅▅▇▇
MetropolisAlgorithm | v0.0.1 | fail | test | ▇▇▇▇▇▇▇▇▇▇▅
MCMCTempering | v0.3.2 | fail | test | ▅▅▇▅▅▅▅▇▇▅▇

EliminateGraphs generates a different random graphs, no longer the same (based of rand, likely unchanged, but I'm guessing some numerical inaccuracy follows).

MetropolisAlgorithm is about numerical accuracy (slightly changed, likely testing with too strict tolerance)

PushVectors is a failed speed test (no longer improving, not a concern)

[IanButterworth]

Great.

Regarding the syntax evolution suggestion from triage, I don't know how to do that. Does someone want to take over here?

[Keno]

goto needs to become a special macro name like VERSION and then use the same mechanism. Claude will be able to do it with that sentence, but if you need me to do it, I can.

[IanButterworth]

Sorry I tried and neither I nor Claude could figure it out. Are there docs?

[Keno]

I'm trying the following prompt:

make `goto` a  conditional keyword like VERSION, then fix the PR by adding a new oldsymbolicgoto HEAD IR form that doesn't have the error check, and using the same mechanism as the `@VERSION` macro to conditionally emit the error-ful version only for new syntax versions.  

I'd put money on that working, but I'll need to check back in the morning.

[Keno]

I pushed Claude's work to https://github.com/JuliaLang/julia/compare/kf/ib/goto_fixes?expand=1. I didn't review it in detail, but looks about right.