Mastodawn

Things I've gathered from reading github issues of the existing #RStats JSON packages (e.g. {jsonlite}) is that:

* Everyone has weird data
* How nesting should be handled is different for every user
* How to handle NA/NULL when writing JSON is different for every user
* How to handle NA/NULL when reading JSON is different for every user.

Summary: JSON + R = prickly beast!

Show thread

coolbutuseless Jul 31, 2023

As a fantastic example of R/JSON issues:

Namibia missing from dataset!

A certified #NoPrize if you can guess why without visiting the issue!!

https://github.com/jeroen/jsonlite/pull/318

#Rstats

Added naStrings option to 'null_to_na()' and precessors by joachim-gassen · Pull Request #318 · jeroen/jsonlite

Tries to address #98 and #314. I was also affected by this corner case (downloading of Google Trends data for Namibia with the {gtrendsR} package failing systematically). As it is likely to regular...

GitHub

Show thread

tyx Jul 31, 2023

@coolbutuseless He-he, and these guys keep mocking Excel after that "gene names converted to dates" story.
On the other hand this bug can be made perfectly reproducible and versioned.

Show thread

Jim Gardner Jul 31, 2023

@tyx @coolbutuseless also this bug doesn’t regularly come up in the work of users at all levels. Eg recently I gave customers a csv of data to work with. They’re gonna open in Excel so I did the same before sending - to make sure they don’t encounter any mangled data. But of course Excel has to turn the value “2 - 4” into a truncated date, silently, and overwrite the original value. I had to go to my upstream R process and change the string to “2 to 4”.

Show thread

tyx Jul 31, 2023

@jimgar @coolbutuseless Sure! But the main point of my sarcasm, other than just cruel laugh I'm always guilty of, is that you can shoot your leg using any language/tool, and R has a pretty sensitive trigger (also a steep learning curve and obscure defaults). Still better than Excel as main data processing tool.

Show thread