ParserCombinator is slow

[CarloLucibello]

I did some perfomance tests for reading a file from a graph using my package FatGraphs;

Pkg.clone("https://github.com/CarloLucibello/FatGraphs.jl")

For comparison I write a graph in a simple text format (Pajek .net). Each of the following function has been run twice to avoid compilations times:

julia> g = Graph(100,1000,seed=1)
Graph{Int64}(100, 1000)

julia> @time writegraph("test.net",g)
  0.243075 seconds (37.02 k allocations: 1.487 MB)
1

julia> @time readgraph("test.net")
  0.003040 seconds (18.54 k allocations: 550.281 KB)
Graph{Int64}(100, 1000)
```
Notice how  perfomance is degraded  when **reading** from a .dot or a .gml file, relying on ParserCombinator:
```julia
julia> @time writegraph("test.dot",g)
  0.001826 seconds (15.23 k allocations: 664.031 KB)
1

julia> @time readgraph("test.dot")
  2.408855 seconds (1.22 M allocations: 49.666 MB, 0.56% gc time)
Graph{Int64}(100, 1000)


julia> @time writegraph("test.gml",g)
  0.001426 seconds (16.83 k allocations: 789.031 KB)
1

julia> @time readgraph("test.gml")
  1.024898 seconds (511.11 k allocations: 18.279 MB, 0.64% gc time)
Graph{Int64}(100, 1000)
```
Probably ParserCombinator has some huge type instability issues. Can those be avoided?

Bye,
Carlo