merge_grammars! interacts weirdly with HerbBenchmarks.PBE_SLIA_.. grammars.

[THinnerichs]

all_rule_grammar = deepcopy(@cfgrammar begin end)
for g in HerbBenchmarks.all_grammars(PBE_SLIA_Track_2019)
           merge_grammars!(all_rule_grammar, g)
end

end up with a very long grammar. I don't know why the rules are not merged properly.

[THinnerichs]

This breaks down to

julia> all_rule_grammar
1: Real = 5


julia> add(x,y) = x+y
add (generic function with 1 method)

julia> add_rule!(all_rule_grammar, :(Real = add(Real, Real)))
1: Real = 5
2: Real = add(Real, Real)


julia> add_rule!(all_rule_grammar, :(Real = add(Real, Real)))
1: Real = 5
2: Real = add(Real, Real)
3: Real = add(Real, Real)


julia> add_rule!(all_rule_grammar, :(Real = add(Real, Real)))
1: Real = 5
2: Real = add(Real, Real)
3: Real = add(Real, Real)
4: Real = add(Real, Real)

[ReubenJ]

This is due to strict equality checks:

HerbGrammar.jl/src/grammar_base.jl

Line 217 in 8a7c25f

if !any(r === rule for rule ∈ g.rules)

I remember now that we did this because the terminals 1 and true were treated as the same, meaning that whichever one was added first was the only rule allowed in the grammar.