Safe memoization during parse graph flattening

[PieterOlivier]

This PR improves the memoization strategy during parse graph flattening. The original proved both incorrect and too slow in the context of the highly ambiguous and cyclic parse forests produced by error recovery.

Special thanks to @arnoldlankamp who came up with the following three heuristics specifying when nodes can be safely cached for memoization:

1. Any non-zero length node that is part of a link with a non-zero length prefix.
2. Any node, which is a part of a prefix for which the above above holds true.
3. Any nullable node that is part of a link for which all prefixes adhere to rule one at some point.

This PR implements a new caching strategy based on these heuristics.

We have gone through great lengths to test the correctness and performance of the code in this PR.
We already had a test suite that tested error recovery on all characters in all Rascal source files in the rascal repo using
two tests: delete the single character and delete all characters until the end-of-line.

We modified these tests to:

• First try to do the error recovery parse without memoization during flattening
• If this succeeded within two seconds, do the same with memoization
• Compare the results

In all cases where the parse without memoization succeeded, the parse with memoization resulted in
exactly the same tree as the tree build without memoization. This in contrast with the old memoization approach where in the same test the result with memoization was often different from the result without memoization.

We have also setup a performance benchmark to compare the speed of "normal" parsing (no error recovery and no ambiguities)
of all Rascal source files in the rascal repo with the speed before this PR. We could not find any speed differences between
the two version, so we are confident this PR does not degrade performance of "normal" parses.

[codecov]

Codecov Report

Attention: Patch coverage is 47.84314% with 133 lines in your changes missing coverage. Please review.

Project coverage is 49%. Comparing base (8c405fc) to head (d647f70).
Report is 254 commits behind head on feat/error-recovery.

Files with missing lines	Patch %	Lines
src/org/rascalmpl/library/util/ErrorRecovery.java	0%	66 Missing ⚠️
...c/org/rascalmpl/parser/gtd/result/struct/Link.java	71%	11 Missing and 9 partials ⚠️
...rg/rascalmpl/parser/util/ParseStateVisualizer.java	0%	15 Missing ⚠️
src/org/rascalmpl/parser/util/DebugUtil.java	0%	14 Missing ⚠️
src/org/rascalmpl/util/visualize/dot/DotGraph.java	0%	9 Missing ⚠️
...pl/parser/gtd/result/out/DefaultNodeFlattener.java	90%	2 Missing and 1 partial ⚠️
...ascalmpl/parser/gtd/result/out/INodeFlattener.java	50%	1 Missing and 1 partial ⚠️
...ser/gtd/result/out/ListContainerNodeFlattener.java	91%	2 Missing ⚠️
...pl/parser/gtd/result/out/SkippedNodeFlattener.java	0%	0 Missing and 1 partial ⚠️
...ser/gtd/result/out/SortContainerNodeFlattener.java	92%	1 Missing ⚠️

Additional details and impacted files

@@                  Coverage Diff                   @@
##             feat/error-recovery   #2136    +/-   ##
======================================================
  Coverage                     49%     49%            
+ Complexity                  6619    6572    -47     
======================================================
  Files                        687     696     +9     
  Lines                      61218   61043   -175     
  Branches                    8874    8910    +36     
======================================================
+ Hits                       30369   30383    +14     
+ Misses                     28620   28410   -210     
- Partials                    2229    2250    +21     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[DavyLandman]

Hi @arnoldlankamp, your comments on #2100 really helped out. Do you have some time to take a look at this? Thanks.

ps. @PieterOlivier I think this means you should close #2100 ?

[PieterOlivier]

ps. @PieterOlivier I think this means you should close #2100 ?

Definitely, I have done so now.

[arnoldlankamp]

Hi @arnoldlankamp, your comments on #2100 really helped out. Do you have some time to take a look at this? Thanks.

I'll have a look at it as soon as I'm able.