[BUG] attr(data, "code") is not reactive

[gogonzo]

Our list of datasets with "code" attribute is buggy. Each dataset is reactive but code don't change when the data changes. It means that filter-panel code changes are not reflected in the "Show R code".

Possible solutions:

• data should be a formal class (can be S3) with reactive code. Should be easy to initialise. I propose to refactor teal.data classes to be something which holds data+code+keys only as all additional methods are not necessary in modules. This means that cdisc_data(...) can be passed also to the modules. Other scenario is to create a new class (e.g. module_data)
• attr(data, "code") can be a reactive()
• instead list or reactive datasets we can have reactive list of datasets.

[Polkas]

In my opinion data with preprocessing code which is static, should be independent of FilterPanel.
The filter panel code could be added in the module with method like filter_panel_api$get_expr().

We should think about passing the data as unfiltered one to module, and the filter_panel_api (available in the module) will be filtering this data and returning the needed expr for the quosure.

# Sth like
tm_x_srv <- function(data[unfiltered], filter_panel_api) {
    data_filtered <- filter_panel_api$apply_filtering(data)
    quosure <- new_quosure(data) %>% eval_code(filter_panel_api$get_expr())
}

regarding "so you have a separate copy of the data for each module", we could reuse already available data sth like
https://www.tutorialspoint.com/design_pattern/flyweight_pattern.htm
In practice it could be even the r memoise.
if one module ask for filtered data it is cached and the second module (no filtering changes made) use this cached data

@nikolas-burkoff

[donyunardi]

After installing the packages properly, I am able to reproduce it.
Since this bug is related to our quosure update, we will have to prioritize this within this increment.
Adding discussion label.