Change parse_row! to a generated function to avoid dynamic dispatch

[nickrobinson251]

• closes Avoid type instability/dynamic dispatch for parsing values #22
• relies on Remove EOL Options (rely on xparse correctly parsing up to the invalid delim) #28

currently this is a bit gross... tbh the generic parse_row! function is actually quite nice (for a generated function), but the Transformers one is pretty gross right now. And in general needs a bit of clean-up. But i think something like this is the way to go.

[codecov-commenter]

Codecov Report

Merging #23 (bee93aa) into main (99ca5ee) will increase coverage by 2.19%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##             main      #23      +/-   ##
==========================================
+ Coverage   95.48%   97.67%   +2.19%     
==========================================
  Files           2        3       +1     
  Lines         177      258      +81     
==========================================
+ Hits          169      252      +83     
+ Misses          8        6       -2     

Impacted Files	Coverage Δ
src/parsing.jl	95.60% <100.00%> (+0.48%)	⬆️
src/transformers_parsing.jl	100.00% <100.00%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 99ca5ee...bee93aa. Read the comment docs.

[nickrobinson251]

Some intial timings (on files with ~250,000 rows, although of course we're currently only parsing up the end of transformers data, so ~90% of that)

on main

julia> using PowerFlowData

julia> @time net = parse_network("test/realistic29.raw");
  5.072275 seconds (49.92 M allocations: 1.661 GiB, 5.07% gc time, 27.00% compilation time)

julia> @time net = parse_network("test/realistic29.raw");
  3.600963 seconds (48.55 M allocations: 1.593 GiB, 2.00% gc time)

julia> @time net = parse_network("test/realistic30.raw");
  3.478833 seconds (43.96 M allocations: 1.441 GiB, 4.28% gc time)

julia> @time net = parse_network("test/realistic30.raw");
  3.369844 seconds (43.96 M allocations: 1.441 GiB, 4.23% gc time)

on this branch

(at 3146cb0)

julia> using PowerFlowData

julia> @time net = parse_network("test/realistic29.raw");
  0.363916 seconds (128.33 k allocations: 55.870 MiB, 15.86% gc time, 44.51% compilation time)

julia> @time net = parse_network("test/realistic29.raw");
  0.226292 seconds (444 allocations: 47.671 MiB, 2.50% gc time)

julia> @time net = parse_network("test/realistic30.raw");
  0.211939 seconds (444 allocations: 47.608 MiB, 3.81% gc time)

julia> @time net = parse_network("test/realistic30.raw");
  0.214129 seconds (444 allocations: 47.608 MiB, 3.66% gc time)

[nickrobinson251]

i think 444 allocations seems a lot more reasonable

Here's how many come from just the constructors

julia> for T in fieldtypes(Network)[2:end]
           @show T fieldcount(T)
           @btime $T(0)
           println("--")
       end
T = Buses
fieldcount(T) = 11
  2.370 μs (27 allocations: 1.50 KiB)
--
T = Loads
fieldcount(T) = 12
  2.590 μs (29 allocations: 1.66 KiB)
--
T = Generators
fieldcount(T) = 20
  4.093 μs (46 allocations: 2.89 KiB)
--
T = Branches
fieldcount(T) = 17
  3.596 μs (40 allocations: 2.48 KiB)
--
T = Transformers
fieldcount(T) = 73
  14.476 μs (152 allocations: 10.02 KiB)
--

[nickrobinson251]

some ideas for when i have another go at refactoring the Transformers bit of this:

• split the code so we only check is_t2 once, e.g. build up the expressions for both sides of that branch, rather than branching multiple times
• could we move out of the generated func all the code that isn't just pulling out the fieldtypes? Can we e.g. pull the types out into a Tuple (T1, T2, T3, ...) and then work with that (I suppose that "work with that" bit might still need to be generated... hmm -- invesitgate something like this anyway)

also, i think these are the cases that appear handle:

1. parse cell
2. parse end cell (and handle EOL comments)
3. if t2; do 2; else 1
4. if t2; fill with missing; else do 1
5. if t2; fill with missing; else do 2

[nickrobinson251]

on this branch with the simplified code (at 08d6278)

julia> using PowerFlowData

julia> @time net = parse_network("test/realistic29.raw");
  0.316417 seconds (128.28 k allocations: 55.867 MiB, 3.57% gc time, 33.23% compilation time)

julia> @time net = parse_network("test/realistic29.raw");
  0.212118 seconds (444 allocations: 47.671 MiB, 4.25% gc time)

julia> @time net = parse_network("test/realistic30.raw");
  0.222221 seconds (444 allocations: 47.608 MiB, 3.28% gc time)

julia> @time net = parse_network("test/realistic30.raw");
  0.207607 seconds (444 allocations: 47.608 MiB)