dazdiff

dazdiff compares two tables value by value. Its one main function, diff_table(), answers one of two questions:

• Does this table match the benchmark? (mode = "benchmark", the default.) The benchmark's column types are the truth. The tested table's columns are cast to them, and its extra columns are dropped.
• Are these two tables equal? (mode = "equal".) The tables are peers: every column must be in both, with the same type, and nothing is cast.

The result is a data.table with one row per compared value. Each row traces back to its source row in both tables.

Installation

Install the released version from CRAN:

install.packages("dazdiff")

Or the development version from GitHub:

# install.packages("pak")
pak::pak("pedrobtz/dazdiff")

Comparing against a benchmark

A benchmark table, and the same data after a round trip through storage. On the way, the column names changed case, the ids became text, units came back as a double, a load timestamp was added, one row was lost and one was added:

library(dazdiff)

benchmark <- data.frame(
  id     = c(1L, 2L, 3L, 4L),
  region = c("north", "south", "east", "west"),
  amount = c(100, 200, 300, 400),
  units  = c(10L, 20L, 30L, 40L)
)
stored <- data.frame(
  ID        = c("1", "2", "3", "5"),
  Region    = c("north", "south", "east", "north"),
  Amount    = c(100, 200.5, 300, 500),
  Units     = c(10.4, 20, 30, 50),
  loaded_at = "2026-09-29"
)

d <- diff_table(benchmark, stored, by = c("id", "region"), measures = "numeric+integer")
#> Dropped column(s) of `y` that the benchmark `x` does not have: loaded_at.
#> Truncated 1 value(s) while casting to the reference types:
#>   Units: 1 (10.4 -> 10)
d
#> Key: <ID, REGION, metric>
#>        ID REGION metric row_id_x row_id_y value_x value_y  diff abs_diff    rel_diff  status
#>     <int> <char> <char>    <int>    <int>   <num>   <num> <num>    <num>       <num>  <fctr>
#>  1:     1  north AMOUNT        1        1     100   100.0   0.0      0.0 0.000000000    same
#>  2:     1  north  UNITS        1        1      10    10.0   0.0      0.0 0.000000000    same
#>  3:     2  south AMOUNT        2        2     200   200.5   0.5      0.5 0.002493766 changed
#>  4:     2  south  UNITS        2        2      20    20.0   0.0      0.0 0.000000000    same
#>  5:     3   east AMOUNT        3        3     300   300.0   0.0      0.0 0.000000000    same
#>  6:     3   east  UNITS        3        3      30    30.0   0.0      0.0 0.000000000    same
#>  7:     4   west AMOUNT        4       NA     400      NA    NA       NA          NA  only_x
#>  8:     4   west  UNITS        4       NA      40      NA    NA       NA          NA  only_x
#>  9:     5  north AMOUNT       NA        4      NA   500.0    NA       NA          NA  only_y
#> 10:     5  north  UNITS       NA        4      NA    50.0    NA       NA          NA  only_y

Reading the result:

• status is same, changed, only_x (the row is missing from the tested table) or only_y (the row exists only there).
• row_id_x and row_id_y are the row numbers in the two inputs.
• diff is value_y - value_x; abs_diff is its size, and rel_diff the size relative to the larger of the two values.
• Column names are normalized in both tables: upper case, with other characters turned into _. That is why id and ID pair up.
• units was cast to the benchmark's integer type. Its 10.4 was truncated to 10, which is announced in a message and recorded, with the rows it touched, in attr(d, "truncated"). Pass truncate = FALSE to refuse such casts instead.

diff_summary() condenses the result per measure:

diff_summary(d)
#>    metric     n n_same n_changed n_only_x n_only_y max_abs_diff mean_abs_diff      rmse max_rel_diff
#>    <char> <int>  <int>     <int>    <int>    <int>        <num>         <num>     <num>        <num>
#> 1: AMOUNT     5      2         1        1        1          0.5     0.1666667 0.2886751  0.002493766
#> 2:  UNITS     5      3         0        1        1          0.0     0.0000000 0.0000000  0.000000000

diff_report() draws the whole comparison as text: the columns compared, how the rows of the two tables align, and how much each measure differs:

This is the beginning of the README. Read the rest at the original. Read the full README ↗