Fix dropal inline

[PalumboN]

Fix #256
Fix #230

Now the compilation of BGMeteorBoard

DROpalCompiler recompileClass: BGMeteorBoard except: #().

finish on all (possible) cases:

1. hasEastOrWestIsland:
Error: Non-local return on inlined method breaks the semantics
2. northIslandsFor:row:
Error: Non-local return on inlined method breaks the semantics
3. findIsland:
Error: Non-local return on inlined method breaks the semantics
4. fillMaskStartingAt:stoppingAbove:ifFoundEnough:
Error: Non-local return on inlined method breaks the semantics
5. hasSouthIsland:
Error: Non-local return on inlined method breaks the semantics
6. islandsFor:
Error: Non-local return on inlined method breaks the semantics

and give these results:

[BenchmarkGameSuite new benchMeteor] bench.  
"ORIGINAL:         35 iterations in 5 seconds 88 milliseconds. 6.879 per second"
"DRUID (noinline): 10 iterations in 5 seconds 133 milliseconds. 1.948 per second"
"DRUID (inline):   12 iterations in 5 seconds 111 milliseconds. 2.348 per second"

We should profile the causes of:

• 30% slowdown from the original Opal to DROpal without inlines
• 20% speedup from DROpal without inlines to with them

what could be related to:

• Temp vector overhead [EPIC] Optimize the temp vectors #251
• Accessing external instance variables New bytecode for access external inst var #259
• JIT incompatibilities Check if the JIT compiles inlined methods #258

[guillep]

Some small questions!

[guillep]

Why dis?

[PalumboN]

DRError does not exist 🤷 I checked the other validations and use Error. It looks like we didn't hit this problem for a long time

[guillep]

3? 5? 7?

[PalumboN]

[guillep]

Really? even in the presence of temp vectors?

[PalumboN]

Yes, but during the compilation we need to track the def-use chain in the SSA. If not, variables used inside a block could be "dead-coded" before inlining.

The problem was that the implementation kept track a lot of trash (values that were not used), adding noise to the debugging.

Anyway, we need to revisit and re-think these decisions when optimizing temp vectors
#231
#251