WIP: rewrite gap_to_julia

[fingolfin]

TODO:

• add more test cases esp. for the various fixes
  e.g. converting the same (identical) objects to multiple
  output types should be working correctly now
• identify issues this patch resolves: Flaw in recursive conversion in gap_to_julia #275, Optimize Vector{GapObj} and some other similar cases #736
• update docs
• update CHANGES,md
• ...

[codecov]

Codecov Report

Merging #777 (7b68aa5) into master (6bafdba) will increase coverage by 2.57%.
The diff coverage is 96.74%.

❗ Current head 7b68aa5 differs from pull request most recent head ea7e465. Consider uploading reports for the commit ea7e465 to get more accurate results

@@            Coverage Diff             @@
##           master     #777      +/-   ##
==========================================
+ Coverage   75.73%   78.31%   +2.57%     
==========================================
  Files          51       46       -5     
  Lines        4159     3652     -507     
==========================================
- Hits         3150     2860     -290     
+ Misses       1009      792     -217     

Files Changed	Coverage
src/gap_to_julia.jl	96.66%
src/constructors.jl	100.00%

[ThomasBreuer]

I do not understand this statement.
Currently we have

julia> GAP.gap_to_julia(GAP.evalstr("[1,Z(2),true]"))
3-element Vector{Any}:
    1
     GAP: Z(2)^0
 true

and

julia> Vector{GAP.Obj}(GAP.evalstr("[1,Z(2),true]"))
3-element Vector{Union{Bool, GAP.FFE, GAP.GapInt}}:
    1
     GAP: Z(2)^0
 true

Shall the latter be forbidden?

[fingolfin]

So we already discussed this on gather, but I want to record what I said here for others / myself in the future:

Indeed the text is not phrased correctly. What I wanted to express is that I don't see how to handle a request like this properly:
Vector{Union{String,Int,Symbol}}(GAP.evalstr("[1, \"x\"]"))

In principle this could be dealt with, in that there are sensible conversion results of the requested type for this, namely [1, "x"] and [1, :x]. But then we need to decide which of them to return? Perhaps do it arbitrarily? Or introduce some kind of priority between certain types (e.g "if both Symbol and String are requested and possible, prefer Symbol"), but then how to extend that to all cases (We could be asked to produce a Vector{Union{Vector{Vector{Int}}, Matrix{Int}, fmpz_mat, fmpq_mat} ...).

In contrast, conversion to Vector{GAP.Obj} above works, despite GAP.Obj being a union type, because all the elements of the GAP list already had a type in that Union. So in that case the "trivial conversions" rule applies, i.e., asking to convert an object x::T to type S always succeeds and returns x whenever S is either T or a supertype of T.

[ThomasBreuer]

As far as I understand, in line 46 above, it is stated that union types are not allowed.