Fix how quotes are handled in guess phase

[davidanthoff]

Fixes #97.

The problem with the old version is that the whole guess phase breaks down if there are line feeds inside quotes in the rows that are used to guess the column types.

I want to trigger code coverage first to get a sense what extra tests I need to add.

@shashi Alright, now ready to be reviewed.

[codecov-io]

Codecov Report

Merging #98 into master will increase coverage by 0.92%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master      #98      +/-   ##
==========================================
+ Coverage   74.76%   75.68%   +0.92%     
==========================================
  Files          11       11              
  Lines        1153     1197      +44     
==========================================
+ Hits          862      906      +44     
  Misses        291      291

Impacted Files	Coverage Δ
src/util.jl	74.13% <100%> (+8.75%)	⬆️
src/csv.jl	78.99% <100%> (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 e34a12a...aa6993f. Read the comment docs.

[shashi]

Is there any hope of using the existing quoted String token parser with includenewline=true? I think it will be awesome if we can use the same code since this code is non trivial.

I think one can just parse quoted stringtoken fields until a field ends exactly at a new line. There's no need to getrowend or even quotedsplit.

[davidanthoff]

Yeah, probably. I started writing getrowend, thinking it would be trivial. Then I thought of more and more corner situations, addressed them, and ended up with this PR... At that point I also felt that there must be a better way to reuse existing stuff.

I think one can just parse quoted stringtoken fields until a field ends exactly at a new line. There's no need to getrowend or even quotedsplit.

That sounds very plausible. I'm not sure when I'll have time again to get to that, though... Any chance you might take a stab at this?

[davidanthoff]

@shashi what should we do about this? Could we maybe merge it for now? It seems to fix a pretty clear bug for now. I agree it is not the most elegant way to do this, and it would be nicer to use the existing machinery for that, but right now from my point of view it would be preferable to fix this bug, and then we can see if someone has time later to make the code more elegant?

[shashi]

Sounds good! :-)