Sparse Notes on {rtables} Internals
Davide Garolini
2024-04-29
Source:vignettes/dev-guide/dg_notes.Rmd
dg_notes.RmdDisclaimer
This is a collection of notes divided by issues and it is a working document that will end up being a developer vignette one day.
section_div notes
Everything in the layout is built over split objects, that reside in
00_tabletrees.R. There section_div is defined
internally in each split object as child_section_div and
assigned to NA_character as default. This needs to be in
all split objects that need to have a separator divisor. Object-wise,
the virtual class Split contains section_div
and it has the following sub-classes. I tagged with “X” constructor that
allows for section_div to be assigned to a value different
than NA_character, and "NX" otherwise.
## Loading required package: formatters## Loading required package: magrittr##
## Attaching package: 'rtables'## The following object is masked from 'package:utils':
##
## str
getClass("Split")## Virtual Class "Split" [package "rtables"]
##
## Slots:
##
## Name: payload name split_label
## Class: ANY character character
##
## Name: split_format split_na_str split_label_position
## Class: FormatSpec character character
##
## Name: content_fun content_format content_na_str
## Class: listOrNULL FormatSpec character
##
## Name: content_var label_children extra_args
## Class: character logical list
##
## Name: indent_modifier content_indent_modifier content_extra_args
## Class: integer integer list
##
## Name: page_title_prefix child_section_div
## Class: character character
##
## Known Subclasses:
## Class "CustomizableSplit", directly
## Class "AllSplit", directly
## Class "VarStaticCutSplit", directly
## Class "VarDynCutSplit", directly
## Class "VAnalyzeSplit", directly
## Class "CompoundSplit", directly
## Class "VarLevelSplit", by class "CustomizableSplit", distance 2
## Class "MultiVarSplit", by class "CustomizableSplit", distance 2
## Class "RootSplit", by class "AllSplit", distance 2
## Class "ManualSplit", by class "AllSplit", distance 2
## Class "CumulativeCutSplit", by class "VarStaticCutSplit", distance 2
## Class "AnalyzeVarSplit", by class "VAnalyzeSplit", distance 2
## Class "AnalyzeColVarSplit", by class "VAnalyzeSplit", distance 2
## Class "AnalyzeMultiVars", by class "CompoundSplit", distance 2
## Class "VarLevWBaselineSplit", by class "VarLevelSplit", distance 3
# Known Subclasses:
# ? Class "CustomizableSplit", directly # vclass used for grouping different split types (I guess)
# Class "AllSplit", directly # NX
# Class "VarStaticCutSplit", directly # X via make_static_cut_split
# Class "VarDynCutSplit", directly # X
# Class "VAnalyzeSplit", directly # X
# ? Class "CompoundSplit", directly # Used only for AnalyzeMultiVars (maybe not needed?)
# Class "VarLevelSplit", by class "CustomizableSplit", distance 2 # X
# Class "MultiVarSplit", by class "CustomizableSplit", distance 2 # X
# Class "RootSplit", by class "AllSplit", distance 2 # NX
# Class "ManualSplit", by class "AllSplit", distance 2 # X
# Class "CumulativeCutSplit", by class "VarStaticCutSplit", distance 2 # X via make_static_cut_split
# Class "AnalyzeVarSplit", by class "VAnalyzeSplit", distance 2 # Virtual
# Class "AnalyzeColVarSplit", by class "VAnalyzeSplit", distance 2 # X
# Class "AnalyzeMultiVars", by class "CompoundSplit", distance 2 # X
# Class "VarLevWBaselineSplit", by class "VarLevelSplit", distance 3 # NXThis can be updated only by related layout functions. The most
important, that are covered by tests are analyze and
split_rows_by.
Now it is relevant to understand where this information is saved in
the table object built by build_table. To do that we need
to see where it is present and how it is assigned. Let’s go back to
00tabletree.Rand look for
trailing_section_div. As classes definitions goes, you will
notice from the search that trailing_section_div is present
in the virtual classes TableRow and
VTableTree. In the following is the class hierarchy that
makes `trailing_section_div:
getClass("TableRow")## Virtual Class "TableRow" [package "rtables"]
##
## Slots:
##
## Name: leaf_value var_analyzed label
## Class: ANY character character
##
## Name: row_footnotes trailing_section_div level
## Class: list character integer
##
## Name: name col_info format
## Class: character InstantiatedColumnInfo FormatSpec
##
## Name: na_str indent_modifier table_inset
## Class: character integer integer
##
## Extends:
## Class "VLeaf", directly
## Class "VTableNodeInfo", directly
## Class "VNodeInfo", by class "VLeaf", distance 2
##
## Known Subclasses: "DataRow", "ContentRow", "LabelRow"
# Extends:
# Class "VLeaf", directly
# Class "VTableNodeInfo", directly
# Class "VNodeInfo", by class "VLeaf", distance 2
#
# Known Subclasses: "DataRow", "ContentRow", "LabelRow"
getClass("VTableTree")## Virtual Class "VTableTree" [package "rtables"]
##
## Slots:
##
## Name: children rowspans labelrow
## Class: list data.frame LabelRow
##
## Name: page_titles horizontal_sep header_section_div
## Class: character character character
##
## Name: trailing_section_div col_info format
## Class: character InstantiatedColumnInfo FormatSpec
##
## Name: na_str indent_modifier table_inset
## Class: character integer integer
##
## Name: level name main_title
## Class: integer character character
##
## Name: subtitles main_footer provenance_footer
## Class: character character character
##
## Extends:
## Class "VTableNodeInfo", directly
## Class "VTree", directly
## Class "VTitleFooter", directly
## Class "VNodeInfo", by class "VTableNodeInfo", distance 2
##
## Known Subclasses: "ElementaryTable", "TableTree"
# Extends:
# Class "VTableNodeInfo", directly
# Class "VTree", directly
# Class "VTitleFooter", directly
# Class "VNodeInfo", by class "VTableNodeInfo", distance 2
#
# Known Subclasses: "ElementaryTable", "TableTree"Always check the constructors after finding the classes. In the above
case for example, the DataRow and ContentRow
share the constructor, so we do not need to add identical getter and
setters for these two classes but only for the virtual class
TableRow. Different is the story for LabelRow
which needs to be handle differently. Now, to understand why only these
two have this feature, lets see the structure of a table built with
section dividers:
lyt <- basic_table() %>%
split_rows_by("ARM", section_div = "+") %>%
split_rows_by("STRATA1", section_div = "") %>%
analyze("AGE",
afun = function(x) list("Mean" = mean(x), "Standard deviation" = sd(x)),
format = list("Mean" = "xx.", "Standard deviation" = "xx."),
section_div = "~"
)
tbl <- build_table(lyt, DM)
print(tbl)## all obs
## ————————————————————————————————
## A: Drug X
## A
## Mean 33
## ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
## Standard deviation 7
##
## B
## Mean 35
## ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
## Standard deviation 7
##
## C
## Mean 36
## ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
## Standard deviation 9
## ++++++++++++++++++++++++++++++++
## B: Placebo
## A
## Mean 32
## ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
## Standard deviation 6
##
## B
## Mean 32
## ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
## Standard deviation 6
##
## C
## Mean 34
## ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
## Standard deviation 7
## ++++++++++++++++++++++++++++++++
## C: Combination
## A
## Mean 36
## ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
## Standard deviation 7
##
## B
## Mean 34
## ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
## Standard deviation 6
##
## C
## Mean 34
## ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
## Standard deviation 6## [1] "TableTree"
## attr(,"package")
## [1] "rtables"
# methods("trailing_section_div") # to see this please do devtools::load_all()
# [1] trailing_section_div,LabelRow-method
# trailing_section_div,TableRow-method
# trailing_section_div,VTableTree-methodIn the above, we show that trailing_section_div has
methods for TableRow virtual object, LabelRow,
and VTableTree. These three make the whole
section_div structure as the VTableTree is
present in TableTree and ElementaryTable that
are the two main table objects. If these are not
NA_character_ then the section_div is printed
at split divisions. The LabelRow and TableRow
are different as their assignment allows the row-wise modification of
separators. When we have a special case for a ContentRow,
as it is represented as content_table(obj) which is a
one-line ElementaryTable, while label row is turned off.
Please take a moment to check the following setter:
setMethod("section_div<-", "VTableTree", function(obj, value, only_sep_sections = FALSE) {
char_v <- as.character(value)
tree_depths <- unname(vapply(collect_leaves(obj), tt_level, numeric(1)))
max_tree_depth <- max(tree_depths)
stopifnot(is.logical(only_sep_sections))
.check_char_vector_for_section_div(char_v, max_tree_depth, nrow(obj))
# Automatic establishment of intent
if (length(char_v) < nrow(obj)) {
only_sep_sections <- TRUE
}
# Case where only separators or splits need to change externally
if (only_sep_sections && length(char_v) < nrow(obj)) {
if (length(char_v) == 1) {
char_v <- rep(char_v, max_tree_depth - 1) # -1 is the data row
}
# Case where char_v is longer than the max depth
char_v <- char_v[seq_len(min(max_tree_depth, length(char_v)))]
# Filling up with NAs the rest of the tree depth section div chr vector
missing_char_v_len <- max_tree_depth - length(char_v)
char_v <- c(char_v, rep(NA_character_, missing_char_v_len))
# char_v <- unlist(
# lapply(tree_depths, function(tree_depth_i) char_v[seq_len(tree_depth_i)]),
# use.names = FALSE
# )
}
# Retrieving if it is a contentRow (no need for labelrow to be visible in this case)
content_row_tbl <- content_table(obj)
is_content_table <- isS4(content_row_tbl) && nrow(content_row_tbl) > 0
# Main table structure change
if (labelrow_visible(obj) || is_content_table) {
if (only_sep_sections) {
# Only tables are modified
trailing_section_div(tt_labelrow(obj)) <- NA_character_
trailing_section_div(obj) <- char_v[1]
section_div(tree_children(obj), only_sep_sections = only_sep_sections) <- char_v[-1]
} else {
# All leaves are modified
trailing_section_div(tt_labelrow(obj)) <- char_v[1]
trailing_section_div(obj) <- NA_character_
section_div(tree_children(obj), only_sep_sections = only_sep_sections) <- char_v[-1]
}
} else {
section_div(tree_children(obj), only_sep_sections = only_sep_sections) <- char_v
}
obj
})only_sep_sections is a parameter that is used to change
only the separators (between splits) and not the data rows. It is
happening forcefully if set to TRUE, but it is
automatically activated when section_div(tbl) <- char_v
is a character vector of length < nrow(tbl). Notice that
the exception for ContentRow is activated by the switcher
is_content_table. This is because content rows do not have
visible label row. You see that in the main table structure change we
have two blocks depending on only_sep_sections. If
TRUE only the VTableTree are modified leading
to only split section separators to be modified. Also consider looking
at section_div getter and tests in
test-accessors.R to have more insights on the structure.
Also to understand exactly how this is bound to output, please check the
result of make_row_df() for the column
trailing_sep. Indeed, an alternative and iterative method
is used by make_row_df to retrieve the information about
the separators for each table row. Being it a trailing separator by
definition, we added header_section_div as a function and a
parameter of basic_table, so to possibly add an empty line
after the header (e.g. header_section_div(tbl) = " "). This
is not a trailing separator, but it is a separator that is added after
the header. To close the circle, please check how
trailing_sep and header_section_div is
propagated and printed/used in formatters::toString.