Package {sas2r}

sas2r: Translate 'SAS' Programs to R with Dependency-Aware Migration

Description

Translates 'SAS' programs and projects into R code, orders connected programs by their dependencies, and executes generated code. Compares generated datasets with supplied reference outputs and reports unresolved translation and validation findings. Rule-based translation works offline without a 'SAS' installation. Optional language model translation, review, and repair use a user-configured provider through 'ellmer' and may require an account and API credentials.

Author(s)

Songyue Chen

See Also

Useful links:

• https://github.com/songyue-chen/sas2r

• Report bugs at https://github.com/songyue-chen/sas2r/issues

Analyze dataset row order independently of content equivalence

Description

Analyze dataset row order independently of content equivalence

Usage

analyze_output_order(reference, candidate, pairs, context = NULL)

Arguments

Value

A list containing order analysis metadata.

Examples

ref <- data.frame(id = 1:3, v = c("a", "b", "c"))
cand <- data.frame(id = c(3, 1, 2), v = c("c", "a", "b"))
pairs <- tibble::tibble(reference_row = 1:3, candidate_row = c(2L, 3L, 1L))

# Content is equivalent; only the row order differs.
res <- analyze_output_order(ref, cand, pairs,
  context = list(order_contract = list(vars = "id", descending = FALSE)))
res$order_equivalent
res$reason

Apply a SAS format

Description

apply_format() applies a compiled format catalog entry (as written to a bundle's ⁠_sas2r_formats.R⁠): exact values first, then ranges (each a list with lo, hi, label), then other for anything unmatched. Missing input stays missing unless the format defines other. sas_put() is PUT(x, fmt.) and does the same. A format name is resolved in the bundle's .sas2r_formats catalog. Names are case-insensitive, with an optional trailing period. Character names retain their $ prefix; numeric and character formats are distinct. Width and decimal specifications are unsupported and raise an error rather than silently changing the requested formatting. An unknown name is an error. Startup loads the catalog before programs and macro functions run; individual formats are not separate R variables.

Usage

apply_format(
  x,
  fmt,
  catalog = get0(".sas2r_formats", envir = parent.frame(), inherits = TRUE)
)

sas_put(
  x,
  fmt,
  catalog = get0(".sas2r_formats", envir = parent.frame(), inherits = TRUE)
)

Arguments

Value

A character vector.

See Also

Other runtime helpers: chr_cmp(), lib_delete(), lib_exists(), lib_members(), lib_read(), sas2r_fold_names(), sas2r_libname_assign(), sas2r_resolve_registry(), sas2r_source_include(), sas_compress(), sas_display(), sas_if_else(), sas_merge(), sas_min(), sas_missing(), sas_round(), sas_sort(), sas_sum()

Examples

sex <- list(values = c(M = "Male", F = "Female"), other = "Unknown")
apply_format(c("M", "F", "X", NA), sex)
age <- list(ranges = list(list(lo = 0, hi = 17, label = "<18"),
                          list(lo = 18, hi = 200, label = "18+")))
sas_put(c(5, 40, NA), age)
sas_put(c("M", "F"), "$SEX.", catalog = list("$sex" = sex))

Serialize a diff digest to JSON

Description

Serialize a diff digest to JSON

Usage

as_digest_json(digest)

Arguments

Value

A JSON string representing the diff digest.

Examples

sas <- data.frame(usubjid = c("01-001", "01-002"), aval = c(1.0, 2.0))
r <- data.frame(usubjid = c("01-001", "01-002"), aval = c(1.0, 2.5))
digest <- diff_digest(compare_datasets(sas, r, keys = "usubjid"))

# The redacted JSON payload that may cross to an LLM.
json <- as_digest_json(digest)
substr(json, 1, 60)

Compare values the way SAS does

Description

SAS comparison of possibly-missing values. Character operands ignore trailing blanks, and a missing character value ("" or NA) sorts below every non-missing one; numeric missing sorts below every number.

Usage

chr_cmp(a, b, op = NULL)

Arguments

Details

Supply an explicit operator in Boolean conditions. Omitting op returns an ordering result: equality is 0, which R treats as false. To test equality, use chr_cmp(a, b, op = "==").

Value

With op NULL, an integer vector of -1, 0, 1 (missing equals missing). With an operator, the logical SAS would produce – never NA.

See Also

Other runtime helpers: apply_format(), lib_delete(), lib_exists(), lib_members(), lib_read(), sas2r_fold_names(), sas2r_libname_assign(), sas2r_resolve_registry(), sas2r_source_include(), sas_compress(), sas_display(), sas_if_else(), sas_merge(), sas_min(), sas_missing(), sas_round(), sas_sort(), sas_sum()

Examples

chr_cmp(c("A", NA, "B  "), c("A", NA, "B"))
chr_cmp(c("A", NA, "B  "), c("A", NA, "B"), op = "==")
chr_cmp(c(61, 70, NA), c(61, 60, NA), op = "==")
chr_cmp(c(1, NA, 3), c(1, 2, 2), op = "<")

Compare aligned outputs under SAS-aware tolerance rules

Description

Compare aligned outputs under SAS-aware tolerance rules

Usage

compare_aligned_outputs(
  reference,
  candidate,
  target,
  context = NULL,
  profile = compare_profile(),
  hashes = list(),
  resource_state = "complete",
  max_examples = OUTPUT_EVIDENCE_MAX_EXAMPLES
)

Arguments

Value

An object of class sas2r_comparison_report.

Examples

ref <- data.frame(usubjid = c("01-001", "01-002"), aval = c(1.0, 2.0))
cand <- data.frame(usubjid = c("01-001", "01-002"), aval = c(1.0, 2.5))
target <- list(target_id = "adam.adsl", logical_dataset = "adam.adsl",
               role = "output", contributing_unit_ids = integer())

report <- compare_aligned_outputs(ref, cand, target = target)
report$report_id
report$examples

Compare two datasets under SAS-aware tolerance rules

Description

A thin wrapper over the shared cell comparator diff_aligned_cells(): rows pair by the given keys (or in row order when keys is NULL), and every cell verdict comes from the same loop the output-review engine uses.

Usage

compare_datasets(base, comp, profile = compare_profile(), keys = profile$keys)

Arguments

Value

An object of class sas2r_comparison.

Examples

sas <- data.frame(usubjid = c("01-001", "01-002", "01-003"),
                  aval = c(1.0, 2.0, 3.0))
r <- data.frame(usubjid = c("01-001", "01-002", "01-003"),
                aval = c(1.0, 2.0, 3.0 + 1e-10))

# A floating-point difference inside tolerance is not a mismatch.
cmp <- compare_datasets(sas, r, keys = "usubjid")
passed(cmp)

# A genuine difference is reported.
r$aval[3] <- 3.5
passed(compare_datasets(sas, r, keys = "usubjid"))

Build a comparison tolerance profile

Description

Every equivalence judgement the comparator makes is parameterized here and nowhere else. Defaults implement the accepted SAS-to-R tolerance rules: SAS null == NA, special missings reported, trailing padding cosmetic, absolute numeric tolerance of 1e-8 and no relative tolerance. Equality uses abs(x-y) <= abs + rel * max(abs(x), abs(y)); relative tolerance can be enabled explicitly for measured quantities.

Usage

compare_profile(
  abs = 1e-08,
  rel = 0,
  sas_null_equals_na = TRUE,
  na_tags = c("report", "ignore", "strict"),
  padding = c("cosmetic", "strict"),
  attrs_cosmetic = c("label", "format.sas"),
  overrides = list(),
  keys = NULL
)

Arguments

Value

An object of class sas2r_profile.

Examples

# Defaults implement the accepted SAS-to-R tolerance rules.
p <- compare_profile()
p$numeric

# Tighten numeric tolerance and make trailing blanks significant.
strict <- compare_profile(abs = 1e-12, rel = 1e-12, padding = "strict")

# Per-variable tolerance override, plus keys used for row matching.
compare_profile(overrides = list(aval = list(abs = 1e-6)),
                keys = c("usubjid", "paramcd"))

Build the redacted diff digest (the only comparison artifact that may cross to an LLM). Names, counts, magnitudes, and enumerated hints only – never cell values, never key values.

Description

Build the redacted diff digest (the only comparison artifact that may cross to an LLM). Names, counts, magnitudes, and enumerated hints only – never cell values, never key values.

Usage

diff_digest(x, label = "dataset")

Arguments

Value

An object of class sas2r_diff_digest.

Examples

sas <- data.frame(usubjid = c("01-001", "01-002"), aval = c(1.0, 2.0))
r <- data.frame(usubjid = c("01-001", "01-002"), aval = c(1.0, 2.5))
cmp <- compare_datasets(sas, r, keys = "usubjid")

# Names, counts and magnitudes only -- never cell values, never key values.
d <- diff_digest(cmp, label = "adsl")
d$vars
d$pattern_hints

Delete explicitly named datasets from a writable library

Description

Removes the stored dataset, rather than just an R object. Missing members produce a message and do not stop deletion of other members. Names are matched case-insensitively across rds, xpt and sas7bdat files.

Usage

lib_delete(libref, members)

Arguments

Details

Only an explicit vector of ordinary SAS member names is supported. Expand a known list into names before calling; prefix lists, ranges and ⁠_ALL_⁠ are not interpreted. Dynamic selection that cannot be resolved must remain an explicit unsupported operation, never a silent no-op.

This helper deletes from the configured write directory. If a requested member also exists in a different read directory, deletion is unsupported: removing the written copy would expose that original again to lib_read(). The whole request is checked before any file is removed, and input library files in a separate read directory are never deleted.

Value

The names of removed members, invisibly.

See Also

Other runtime helpers: apply_format(), chr_cmp(), lib_exists(), lib_members(), lib_read(), sas2r_fold_names(), sas2r_libname_assign(), sas2r_resolve_registry(), sas2r_source_include(), sas_compress(), sas_display(), sas_if_else(), sas_merge(), sas_min(), sas_missing(), sas_round(), sas_sort(), sas_sum()

Examples

local({
  folder <- tempfile("work")
  .sas2r_registry <- list(work = list(path = folder, write = "rds"))
  lib_write(data.frame(x = 1), "work", "scratch")
  lib_delete("work", "scratch")
  unlink(folder, recursive = TRUE)
})

Test whether a library dataset member exists

Description

Uses the same registry and write/read search order as lib_read(), without reading or parsing data. A present unreadable or zero-row dataset exists. Invalid arguments and an unknown library remain errors, not FALSE. Supports ordinary RDS, SAS7BDAT and XPT member files, not SAS views or arbitrary external paths. Existence does not establish readability.

Usage

lib_exists(libref, member)

Arguments

Value

A single logical value.

See Also

Other runtime helpers: apply_format(), chr_cmp(), lib_delete(), lib_members(), lib_read(), sas2r_fold_names(), sas2r_libname_assign(), sas2r_resolve_registry(), sas2r_source_include(), sas_compress(), sas_display(), sas_if_else(), sas_merge(), sas_min(), sas_missing(), sas_round(), sas_sort(), sas_sum()

Examples

local({
  .sas2r_registry <- list(work = list(path = tempdir()))
  lib_exists("work", "not_created")
})

List ordinary dataset members in a configured library

Description

Lists names without reading dataset rows. Uses the same member resolver and write/read precedence as lib_exists() and lib_read(). Duplicate storage formats yield one name. Names and extensions must be resolvable on the local filesystem; case is preserved. Unsupported views, name literals and ranges are not expanded. An unknown library or inaccessible configured directory is an error, distinct from an empty library. A not-yet-created write directory is empty; a missing separate read directory remains a configuration error.

Usage

lib_members(libref)

Arguments

Details

Listing does not authorize deletion of protected inputs: lib_delete() retains its separate-input restriction. This is a runtime helper, not an agent tool or permission to send library contents to a model.

Value

A sorted character vector of ordinary resolvable member names.

See Also

Other runtime helpers: apply_format(), chr_cmp(), lib_delete(), lib_exists(), lib_read(), sas2r_fold_names(), sas2r_libname_assign(), sas2r_resolve_registry(), sas2r_source_include(), sas_compress(), sas_display(), sas_if_else(), sas_merge(), sas_min(), sas_missing(), sas_round(), sas_sort(), sas_sum()

Examples

local({
  folder <- tempfile()
  dir.create(folder)
  .sas2r_registry <- list(work = list(path = folder))
  lib_write(data.frame(id = 1), "work", "scratch")
  members <- lib_members("work")
  lib_delete("work", members)
  unlink(folder, recursive = TRUE)
})

Read and write datasets through librefs

Description

lib_read("lib", "member") reads a dataset from a libref as the registry configures it: the libref's write path is tried first (so a dataset the bundle produced earlier wins), then its read path, for .rds, .sas7bdat, and .xpt in that order. The result carries class sas2r_dataset, whose $ and [[ methods resolve column names case-insensitively, as SAS variable names do. lib_write(df, "lib", "member") writes to the libref's write path in its configured format (rds or xpt), creating the directory on demand.

Usage

lib_read(libref, member, ...)

lib_write(df, libref, member, ...)

Arguments

Details

Exactly one call form is accepted for each. Combined "lib.member" strings, single-argument calls, and ⁠dataset=⁠/⁠table=⁠ aliases fail with the canonical form in the message, so a wrong call in an agent translation becomes repair feedback instead of being silently reinterpreted.

The registry is the .sas2r_registry list a bundle's autoexec.R defines; in a translated program it is loaded by the bootstrap header. To use these helpers interactively, create runtime <- new.env() and source that file with ⁠local = runtime, chdir = TRUE⁠, then call runtime$lib_read() or runtime$lib_write().

Value

lib_read() returns the dataset with class sas2r_dataset; lib_write() returns df, invisibly.

See Also

Other runtime helpers: apply_format(), chr_cmp(), lib_delete(), lib_exists(), lib_members(), sas2r_fold_names(), sas2r_libname_assign(), sas2r_resolve_registry(), sas2r_source_include(), sas_compress(), sas_display(), sas_if_else(), sas_merge(), sas_min(), sas_missing(), sas_round(), sas_sort(), sas_sum()

Examples

local({
  lib <- tempfile("lib"); dir.create(lib)
  .sas2r_registry <- list(
    raw = list(read_path = lib, write_path = lib, engine = "rds", write = "rds")
  )
  lib_write(data.frame(USUBJID = c("01", "02"), AVAL = c(1, 2)), "raw", "dm")
  dm <- lib_read("raw", "dm")
  dm$usubjid          # case-insensitive column access
  unlink(lib, recursive = TRUE)
})

Did a comparison pass within tolerance?

Description

Did a comparison pass within tolerance?

Usage

passed(x)

Arguments

Value

Logical indicating whether the comparison passed.

Examples

sas <- data.frame(usubjid = c("01-001", "01-002"), aval = c(1.0, 2.0))
r <- data.frame(usubjid = c("01-001", "01-002"), aval = c(1.0, 2.0))
passed(compare_datasets(sas, r, keys = "usubjid"))

Print a sas2r translation summary

Description

Print a sas2r translation summary

Usage

## S3 method for class 'sas2r_translation'
print(x, ...)

Arguments

Value

The input x, invisibly.

Print a sas2r transpilation summary

Description

Print a sas2r transpilation summary

Usage

## S3 method for class 'sas2r_transpilation'
print(x, ...)

Arguments

Value

The input x, invisibly.

Define reusable dataset quality requirements

Description

Assign a profile a name in outputs$profiles, then select it with outputs$assertions[["adam.adsl"]]$profile. Target assertions override whole profile fields. These are explicit study requirements, not an automatic CDISC compliance assessment or evidence of SAS equivalence.

Usage

qc_profile(
  required_columns = NULL,
  labels = NULL,
  formats = NULL,
  types = NULL,
  column_order = NULL,
  keys = NULL,
  unique_keys = NULL,
  row_count = NULL,
  min_rows = NULL,
  max_rows = NULL,
  numeric_tolerance = NULL,
  tolerances = NULL
)

Arguments

Value

A plain list containing only explicitly supplied, non-NULL fields. Omitted fields inherit global settings. Explicit FALSE and empty lists override them. Types describe physical R columns: factors do not satisfy character, even when their values compare equal to character data.

Examples

adsl_qc <- qc_profile(
  required_columns = c("USUBJID", "AGE"),
  keys = "USUBJID", unique_keys = TRUE,
  types = c(USUBJID = "character", AGE = "numeric"),
  min_rows = 1, tolerances = list(AGE = list(abs = 0, rel = 0))
)

Read a bounded comparison report from the registry

Description

Read a bounded comparison report from the registry

Usage

read_comparison_report(report_id, registry)

Arguments

Value

An object of class sas2r_comparison_report.

Examples

ref <- data.frame(usubjid = "01-001", aval = 1.0)
cand <- data.frame(usubjid = "01-001", aval = 2.0)
target <- list(target_id = "adam.adsl", logical_dataset = "adam.adsl",
               role = "output", contributing_unit_ids = integer())
report <- compare_aligned_outputs(ref, cand, target = target)

# The registry may be a named list or an environment.
registry <- list()
registry[[report$report_id]] <- report
back <- read_comparison_report(report$report_id, registry)
identical(back$report_id, report$report_id)

Resolve a translated LIBNAME path against the execution root

Description

Registry seed paths are resolved separately by sas2r_resolve_registry().

Usage

sas2r_assignment_path(path, env)

Arguments

Value

A normalized character scalar path. Relative paths use the execution root, or the working directory when no root is set. The target need not exist.

Fold column names to lower case

Description

SAS resolves variable names case-insensitively; R does not. Deterministic translations fold every frame to lower-case names at entry so lower-case source references (aval) find upper-case data columns (AVAL) the way SAS would.

Usage

sas2r_fold_names(df)

Arguments

Value

df with lower-case column names.

See Also

Other runtime helpers: apply_format(), chr_cmp(), lib_delete(), lib_exists(), lib_members(), lib_read(), sas2r_libname_assign(), sas2r_resolve_registry(), sas2r_source_include(), sas_compress(), sas_display(), sas_if_else(), sas_merge(), sas_min(), sas_missing(), sas_round(), sas_sort(), sas_sum()

Examples

names(sas2r_fold_names(data.frame(USUBJID = 1, AVAL = 2)))

Read one library entry from the loaded registry

Description

Read one library entry from the loaded registry

Usage

sas2r_lib_entry(libref)

Arguments

Value

The registry entry as a list, normally with read_path, write_path, engine, and write fields. Signals sas2r_unknown_libref when absent; does not read a dataset or return a data frame.

Find a dataset member without reading its contents

Description

Find a dataset member without reading its contents

Usage

sas2r_lib_member_file(reg, member)

Arguments

Value

A list containing type and path, or NULL for an absent member. The write directory precedes the read directory; formats are tried in RDS, SAS7BDAT, XPT order. Directories are not dataset member files.

Construct a member filename inside a library directory

Description

Construct a member filename inside a library directory

Usage

sas2r_lib_member_path(dir, member, ext)

Arguments

Value

A character scalar filename. Does not read or create the file. Invalid member names signal sas2r_libref_member_error.

Bind or clear a libref at its point of use

Description

The registry a bundle starts with is only a seed – the libraries the configuration declares, plus work. Every LIBNAME the SAS program executes changes the registry where that statement stood, so a libref that is rebound or cleared part-way through behaves at run time the way it does in SAS. sas2r_libname_assign() is ⁠LIBNAME libref '<path>'⁠; sas2r_libname_clear() is ⁠LIBNAME libref CLEAR⁠. Relative assignment paths use .sas2r_execution_root from autoexec.R, initially the source project directory. Registry seed paths still use the bundle folder. Reassigning a known physical library preserves its separate write directory. Newly assigned libraries write under the execution output ⁠libraries/⁠ folder unless write_path is explicitly supplied.

Usage

sas2r_libname_assign(
  libref,
  read_path,
  write_path = read_path,
  engine = "sas7bdat",
  write = "rds"
)

sas2r_libname_clear(libref)

Arguments

Value

sas2r_libname_assign() returns the new registry entry, invisibly; sas2r_libname_clear() returns NULL, invisibly.

See Also

Other runtime helpers: apply_format(), chr_cmp(), lib_delete(), lib_exists(), lib_members(), lib_read(), sas2r_fold_names(), sas2r_resolve_registry(), sas2r_source_include(), sas_compress(), sas_display(), sas_if_else(), sas_merge(), sas_min(), sas_missing(), sas_round(), sas_sort(), sas_sum()

Examples

local({
  .sas2r_registry <- list()
  sas2r_libname_assign("raw", tempdir(), engine = "rds")
  names(.sas2r_registry)
  sas2r_libname_clear("raw")
  names(.sas2r_registry)
})

Signal a runtime library error

Description

Signal a runtime library error

Usage

sas2r_libref_stop(cls, msg, identifiers = character())

Arguments

Value

Does not return normally. Signals an error with classes cls, sas2r_libref_error, error, and condition, and fields message, call and identifiers.

Find the environment holding the runtime library registry

Description

Looks first along the runtime's enclosing environments, then active local call frames. Package helpers exclude the global workspace; standalone helpers use the environment they were explicitly sourced into. Load autoexec.R into a local environment for interactive use.

Usage

sas2r_registry_env()

Value

An environment containing .sas2r_registry, not the registry list. Signals sas2r_no_registry if no registry is loaded.

Resolve a bundle's libref registry against its folder

Description

autoexec.R lists each libref's directories the way a person writes them: an absolute path is used as is, and a relative path means inside the bundle folder. This turns the relative ones into absolute paths against root, creates every write directory, and returns the registry ready for lib_read() and lib_write(). autoexec.R calls it once, after the helpers are loaded; call it yourself only when building a registry by hand.

Usage

sas2r_resolve_registry(registry, root)

Arguments

Details

A ⁠<FILL>⁠ placeholder left in a path – the commented entries autoexec.R carries for a library nothing binds – is refused with the libref named, rather than becoming a directory called ⁠<FILL...>⁠.

Value

The registry, with every read_path and write_path absolute.

See Also

Other runtime helpers: apply_format(), chr_cmp(), lib_delete(), lib_exists(), lib_members(), lib_read(), sas2r_fold_names(), sas2r_libname_assign(), sas2r_source_include(), sas_compress(), sas_display(), sas_if_else(), sas_merge(), sas_min(), sas_missing(), sas_round(), sas_sort(), sas_sum()

Examples

root <- tempfile("bundle"); dir.create(root)
reg <- sas2r_resolve_registry(
  list(work = list(read_path = "work", write_path = "work",
                   engine = "rds", write = "rds")),
  root)
reg$work$write_path      # <root>/work, and the directory now exists

The runtime every translated program carries

Description

Every program sas2r translates calls a small runtime – lib_read(), lib_write(), sas_sort(), sas_merge(), chr_cmp(), the format machinery and the rest. Since sas2r 0.2.0 that runtime is exported package code with a help page per helper, and every bundle still carries its own copy as sas2r-helpers.R, rendered from the same source files (⁠R/runtime-*.R⁠) so the two are identical function for function. A bundle always loads its own runtime copy, so it can run without sas2r installed. Results still depend on input data, R and dependency versions; recheck outputs when those change.

The vignette vignette("runtime-helpers", package = "sas2r") explains how to run programs, how to use the runtime interactively, and the stability policy behind it.

How the runtime reaches a program

One guarded line at the top of each generated program sources autoexec.R into the program's own environment, the way a SAS session runs its autoexec first. Run ⁠Rscript run.R⁠ from the bundle folder to execute programs in dependency order. For individual programs, create runtime <- new.env() and use source("<bundle>/autoexec.R", local = runtime, chdir = TRUE). Source required upstream programs into that same environment before source("<bundle>/programs/adsl.R", local = runtime). autoexec.R is the bundle's counterpart of a SAS autoexec: generated by the translation, maintained by a person afterwards. Its LIBRARIES section says where each libref reads from and writes to (a relative path means inside the bundle folder); its LOAD section sources runtime/sas2r-helpers.R (this runtime), resolves those paths with sas2r_resolve_registry(), and sources ⁠runtime/_sas2r_formats.R⁠ (the compiled format catalog).

To use the runtime at the console, load it into runtime as above and call helpers through that environment, for example runtime$lib_read("adam", "adsl"). Attach sas2r when you want the help pages.

The helpers

Data access

lib_read(), lib_write(), sas2r_libname_assign(), sas2r_libname_clear(), sas2r_resolve_registry(), sas2r_fold_names(); datasets read through lib_read() carry class sas2r_dataset, whose $ and [[ methods resolve column names case-insensitively.

SAS value semantics

chr_cmp(), sas_if_else(), sas_sum() and sas_mean(), sas_min() and sas_max(), sas_round(), sas_length(), sas_substr(), sas_compress(), sas_display(); and two operators the package defines but does not export: a %notin% b (⁠NOT IN⁠ with SAS missing semantics – missing values match missing values in the list; character comparison ignores trailing blanks) and a %+% b (|| concatenation where a missing operand contributes nothing rather than "NA"; SAS pads fixed-width operands, this does not).

Sorting and merging

sas_sort(), sas_merge().

Formats

apply_format(), sas_put().

Includes

sas2r_source_include().

Internal helpers

Generated code calls these; they are not exported, not meant for direct use, and may change with the generator: sas2r_registry_env() (where the .sas2r_registry list lives – lexically where the runtime was loaded, then active local frames; package helpers exclude the global workspace), sas2r_lib_entry() (the registry entry for a libref), sas2r_lib_member_path() (the file a member resolves to, refusing any name that is not a plain member name), sas2r_assignment_path() (resolve a LIBNAME path against the execution root), sas2r_libref_stop() (classed libref errors), split_ds() (split a two-level SAS name after macro substitution).

Versioning and stability

The runtime follows the sas2r version, and every vendored copy states the version that generated it. Existing helper contracts are stable, but correctness fixes may change erroneous results. A helper is retired only after a replacement ships and the old name has warned for at least one minor version. Tests hold the source files, the vendored template, the help pages, and the lint allowlist in sync. A separate sas2r.runtime package is a future option when usage justifies it.

See Also

sas_translate(), sas_write(), sas_code()

Run an included module where the %INCLUDE stood

Description

Each included SAS file is translated once into its own staged module; this executes that module at one include site, into the calling program's own environment, exactly where the ⁠%INCLUDE⁠ stood.

Usage

sas2r_source_include(relative_path, envir = parent.frame())

Arguments

Details

The path is relative to the staged module root, and never to the working directory. The editable bundle sets this to ⁠programs/⁠ in autoexec.R; flat execution snapshots use the nearest ancestor holding autoexec.R. Absolute paths and . or .. components are rejected, and the file the joined path actually resolves to is then required to lie under the root: rejecting the components alone would still let a symlinked directory inside the bundle read outside it, so the confinement is checked on the resolved path, where a hand-edited bundle cannot talk its way out.

Value

The resolved path of the module, invisibly.

See Also

Other runtime helpers: apply_format(), chr_cmp(), lib_delete(), lib_exists(), lib_members(), lib_read(), sas2r_fold_names(), sas2r_libname_assign(), sas2r_resolve_registry(), sas_compress(), sas_display(), sas_if_else(), sas_merge(), sas_min(), sas_missing(), sas_round(), sas_sort(), sas_sum()

Examples

## Not run: 
# inside a translated program, where the SAS source had %include 'inc/prep.sas';
sas2r_source_include("inc/prep.R")

## End(Not run)

Retrieve staged code for a translated program from the selected bundle

Description

Recursively inspects the selected bundle directory and resolves by component ID, relative file path, basename, or integer index.

Usage

sas_code(x, file = 1L)

Arguments

Value

Character string containing the program code.

Examples

sas_dir <- file.path(tempdir(), "sas2r-code-example")
dir.create(sas_dir, showWarnings = FALSE)
writeLines(c("proc sort data=work.dm out=work.dm_sorted;",
             "  by usubjid;",
             "run;"), file.path(sas_dir, "sort.sas"))
res <- sas_translate(sas_dir, out_dir = tempfile("sas2r-out"), execute = FALSE)

# Resolve by index, or by file name / component id.
cat(sas_code(res, 1L))

SAS character functions

Description

sas_compress() is COMPRESS: every character listed in chars is removed literally, including ⁠]⁠ and -. SAS modifiers are unsupported. sas_substr() is SUBSTR; a NULL len reads to the end of the string. sas_length() is LENGTH: trailing blanks do not count, and a missing or blank value has length 1, as in SAS.

Usage

sas_compress(x, chars = " ")

sas_substr(x, pos, len = NULL)

sas_length(x)

Arguments

Value

sas_compress() and sas_substr() return character vectors; sas_length() an integer vector.

See Also

Other runtime helpers: apply_format(), chr_cmp(), lib_delete(), lib_exists(), lib_members(), lib_read(), sas2r_fold_names(), sas2r_libname_assign(), sas2r_resolve_registry(), sas2r_source_include(), sas_display(), sas_if_else(), sas_merge(), sas_min(), sas_missing(), sas_round(), sas_sort(), sas_sum()

Examples

sas_compress("a b c")
sas_substr("abcdef", 2, 3)
sas_length(c("abc  ", "", NA))

Load project configuration (optional by design)

Description

Looks for ⁠_sas2r.yml⁠ upward from start; the SAS2R_CONFIG environment variable overrides discovery. No config file is the primary supported case: built-in defaults let a bare SAS script be scanned, assessed, and translated with no setup at all. Every relative configured path – libraries, macros.search_path, includes.roots, environment.autoexec, migration.execution_order, and output/comparison reference paths – is resolved against the configuration file's own directory, never against the working directory: those roots reach ⁠%include⁠ occurrence identity, which must not depend on where the scan was launched from. One anchoring rule governs these paths, so a ~-prefixed entry is treated as already carrying its own base everywhere rather than in some keys only, because R expands ~ against the user's home directory and prefixing a base onto it would corrupt the path.

Usage

sas_config(path = NULL, start = ".")

Arguments

Details

A caller-supplied configuration list has no file of its own, so sas_project() resolves whatever is still relative against the project root. Both bases are facts about the project's layout, and both travel with the checkout.

YAML boolean settings use true and false. Tokens such as N, Y, yes, and no remain strings, preserving clinical metadata keys. Quote string metadata values that look numeric, such as SAS format "8.".

migration.execution_order optionally lists every executable root program exactly once, in its intended order within one shared WORK session. Reads use the latest preceding completed write; a later writer cannot supply an earlier read. List roots only: called macros, startup and included files retain their existing roles. The list orders the scanned source scope; path and recursive in sas_preflight() or sas_translate() still select that scope. Omitting the order retains inferred dependencies. Changing the order on a previously scanned project requires rescanning the sources.

Value

A sas2r_config object containing libraries, macro_search_path, include_roots, migration, optional normalized llm, output_review, outputs, source, and raw.

Examples

# Read the demo project configuration shipped with the package.
cfg <- sas_config(system.file("examples", "migration-demo", "_sas2r.yml",
                              package = "sas2r"))
class(cfg)
names(cfg$libraries)

# With no explicit file, search from a known project directory.
cfg <- sas_config(start = system.file("examples", "migration-demo", package = "sas2r"))

Display values the way SAS prints them

Description

Numeric missing is ".", character missing is blank, and numbers are never in scientific notation.

Usage

sas_display(x)

Arguments

Value

A character vector.

See Also

Other runtime helpers: apply_format(), chr_cmp(), lib_delete(), lib_exists(), lib_members(), lib_read(), sas2r_fold_names(), sas2r_libname_assign(), sas2r_resolve_registry(), sas2r_source_include(), sas_compress(), sas_if_else(), sas_merge(), sas_min(), sas_missing(), sas_round(), sas_sort(), sas_sum()

Examples

sas_display(c(1, NA, 1e6))
sas_display(c("a", NA))

IF/THEN/ELSE assignment with SAS semantics

Description

A missing condition is false in SAS. The result keeps the class (Date, labelled) of the branch that carries it, which ifelse() would strip.

Usage

sas_if_else(cond, yes, no)

Arguments

Value

A vector of the common length.

See Also

Other runtime helpers: apply_format(), chr_cmp(), lib_delete(), lib_exists(), lib_members(), lib_read(), sas2r_fold_names(), sas2r_libname_assign(), sas2r_resolve_registry(), sas2r_source_include(), sas_compress(), sas_display(), sas_merge(), sas_min(), sas_missing(), sas_round(), sas_sort(), sas_sum()

Examples

sas_if_else(c(TRUE, NA, FALSE), "yes", "no")
sas_if_else(c(TRUE, FALSE), as.Date("2026-01-01"), NA)

Construct an LLM adapter from a project's ⁠llm:⁠ configuration

Description

The supported way to obtain the llm argument that sas_translate() accepts. The adapter is backed by public ellmer APIs for one of the twelve registered provider ids. Credential values may be held in memory by adapters and redactors. Keep adapters private; reports redact registered secret values. Constructing an adapter contacts no network – use sas_llm_probe() to validate connectivity and model capabilities, and sas_llm_models() to see the inventory visible to the configured identity.

Usage

sas_llm(config)

Arguments

Value

A sas2r_llm adapter.

See Also

sas_llm_probe(), sas_llm_models()

Examples

# Constructing an adapter contacts no network.
if (requireNamespace("ellmer", quietly = TRUE)) {
  llm <- sas_llm(list(provider = "anthropic", model = "claude-sonnet-4-6"))
  class(llm)
}

## Not run: 
# Choose one of sas2r's registered providers.
llm <- sas_llm(list(provider = "openai", model = "gpt-4.1"))
res <- sas_translate("path/to/sas", llm = llm)

## End(Not run)

List models visible to the configured provider identity

Description

Uses the provider's read-only inventory route when ellmer exposes one. An unavailable or non-authoritative inventory is reported explicitly and is never represented as an empty list of models. Bedrock inference-profile and custom-runtime selectors require probe-based access validation because the foundation-model inventory does not enumerate them safely.

Usage

sas_llm_models(config)

Arguments

Value

A sas2r_llm_model_inventory record.

Examples

## Not run: 
# Contacts the provider's read-only inventory route; needs credentials.
sas_llm_models(list(provider = "anthropic", model = "claude-sonnet-4-6"))

# Or straight from a project configuration.
sas_llm_models(sas_config("path/to/_sas2r.yml"))

## End(Not run)

Probe LLM connectivity and configured settings with a structured ping

Description

Validates authentication, endpoint reachability, and structured-output support and forwarding of explicit parameters on the configured model. The ping uses the configured output ceiling, or 2048 tokens when reasoning effort is configured, otherwise 32. It does not perform the negative control used by automatic translation startup to detect ignored reasoning settings, and does not populate that startup cache. The probe may invoke provider-managed authentication, including device login. A missing or expired ambient session raises a classed condition naming the command to run instead.

Usage

sas_llm_probe(
  llm,
  max_retries = 2L,
  log_dir = tempdir(),
  on_charge = NULL,
  tier = "cheap",
  can_attempt = NULL,
  usage_budget = NULL,
  audit_context = list()
)

Arguments

Details

The ping carries no tools, so it does not exercise tool calling. A model that answers the probe may still lack tool support; that is caught at the first tool-using phase, not here.

An attempt that reaches the provider is written to the audit log. A call refused before transport – by the capability gate, or by an exhausted budget – returns FALSE without an audit entry. Reaching the durable usage ledger additionally requires usage_budget to carry a ledger_path, as it does when a translation run probes through its own shared budget; a standalone call with no usage_budget gets a transient budget with none, so it is audited but not ledgered.

Value

TRUE when the model answered the ping, FALSE on a definitive nonanswer. Access, authentication, and transport failures are raised as classed conditions with secrets redacted.

See Also

sas_llm(), sas_llm_models()

Examples

## Not run: 
# Sends a structured ping to validate connectivity and model access.
llm <- sas_llm(list(provider = "anthropic", model = "claude-sonnet-4-6"))
sas_llm_probe(llm)

## End(Not run)

Merge two datasets the way MERGE ... BY does

Description

⁠MERGE a b; BY by;⁠ with ⁠IN=⁠ semantics selected by keep: "both" (⁠in_a and in_b⁠), "left" (in_a), "right" (in_b), "left_only", "right_only", "full". Where both datasets carry a non-key column the later dataset's value wins, columns follow statement order, and rows follow SAS BY ordering. Many-to-many keys, and duplicate keys with shared non-key columns, are refused: those cases require SAS observation-by-observation semantics that this join-based helper does not implement. Do not pre-create missing columns on one input when the other input supplies them: that invents an overlap. Preserve genuine overlapping source columns; their unsupported repeated-key semantics must not be silently dropped. Values from an input can be retained within a BY group, but reset to missing at the next BY group. If an input has no row for a key, its unique columns are missing for that key; never fill them from a preceding key.

Usage

sas_merge(
  a,
  b,
  by,
  keep = c("full", "both", "left", "right", "left_only", "right_only"),
  ...
)

Arguments

Value

The merged data frame.

See Also

Other runtime helpers: apply_format(), chr_cmp(), lib_delete(), lib_exists(), lib_members(), lib_read(), sas2r_fold_names(), sas2r_libname_assign(), sas2r_resolve_registry(), sas2r_source_include(), sas_compress(), sas_display(), sas_if_else(), sas_min(), sas_missing(), sas_round(), sas_sort(), sas_sum()

Examples

a <- data.frame(id = c(1, 2), x = c("a", "b"))
b <- data.frame(id = c(2, 3), y = c("B", "C"))
sas_merge(a, b, by = "id", keep = "full")
sas_merge(a, b, by = "id", keep = "both")
events <- data.frame(id = c(1, 1, 2), event = c("a", "b", "c"))
subjects <- data.frame(id = c(1, 2), base = c(10, 20))
sas_merge(events, subjects, by = "id", keep = "left") # three events remain

Row-wise MIN and MAX with SAS missing semantics

Description

Row-wise across the arguments, ignoring missing values; an observation whose arguments are all missing yields missing, and a single argument passes through elementwise – SAS MIN(x) is never a column aggregate.

Usage

sas_min(...)

sas_max(...)

Arguments

Value

A numeric vector of that length.

See Also

Other runtime helpers: apply_format(), chr_cmp(), lib_delete(), lib_exists(), lib_members(), lib_read(), sas2r_fold_names(), sas2r_libname_assign(), sas2r_resolve_registry(), sas2r_source_include(), sas_compress(), sas_display(), sas_if_else(), sas_merge(), sas_missing(), sas_round(), sas_sort(), sas_sum()

Examples

sas_min(c(1, NA, 3), c(2, NA, 1))
sas_max(c(1, NA, 3), c(2, NA, 1))

SAS missing values and Boolean conditions

Description

Character missing is NA or all spaces. Numeric missing is NA (including tagged missing values). A missing or zero numeric condition is false. Character conditions convert to numeric; nonnumeric text is false.

Usage

sas_missing(x)

sas_true(x)

Arguments

Value

A logical vector with no missing values.

See Also

Other runtime helpers: apply_format(), chr_cmp(), lib_delete(), lib_exists(), lib_members(), lib_read(), sas2r_fold_names(), sas2r_libname_assign(), sas2r_resolve_registry(), sas2r_source_include(), sas_compress(), sas_display(), sas_if_else(), sas_merge(), sas_min(), sas_round(), sas_sort(), sas_sum()

Examples

sas_missing(c("", " ", NA, "A"))
sas_true(c(NA, 0, 1, -1))

Inspect a migration and optionally diagnose findings

Description

Scans sources and resolves library bindings with the same project scanner used by sas_translate(). Reads directory listings, not dataset contents. Does not execute programs, read dataset contents, or create output directories, scan caches, or usage ledgers. With a configured LLM, actionable findings can receive one bounded advisory diagnosis. Use diagnose = "off" for offline inspection without constructing an adapter or making a model call. Deterministic support findings are advisory: an AI translation may handle deferred constructs, and runtime checks can still defer supported shapes.

Usage

sas_preflight(
  path,
  out_dir = NULL,
  config = NULL,
  outputs = NULL,
  budget_usd = Inf,
  budget_mode = "stop",
  pricing_source = "catalog",
  pricing_rates = NULL,
  usage_limits = NULL,
  recursive = FALSE,
  max_parallel_translations = NULL,
  diagnose = c("auto", "off"),
  llm = NULL
)

Arguments

Details

With diagnose = "auto", a configured model may receive bounded SAS statements, source paths and findings, subject to budget admission; dataset contents are not sent. Configured adapters use one transport attempt capped at 120 seconds; custom supplied adapters retain their transport contract. Advice can suggest source/configuration corrections or an issue draft for a suspected sas2r bug. It does not edit sources or submit issues.

Configure migration.execution_order to inspect a declared shared-session sequence; see sas_config(). pipeline$order_source reports "configured" or "inferred". A permanent input with no preceding known or possible writer still receives the normal availability check, even if a later program writes it. Unknown macro effects can leave dataset state deferred.

Value

A sas2r_preflight list with sources, point-of-use libraries, inputs (available, missing, unresolved, no_producer, deferred, backward_dependency, created_if_missing, generated, or environment), references, unsupported, scanner findings, shared readiness warnings, outputs, destinations, budget, next_actions, diagnosis, and actual model_calls. Diagnosis failures preserve the static result. Inspection errors retain their original condition with optional advice attached as ⁠$diagnosis⁠. called_macros lists reachable search-path macro definitions and their standalone R and interface-test paths. pipeline reconciles every scanned file with the translation schedule and bundle execution roles, listing intentional exclusions and blocking gaps. Also includes status, configured_libraries, schedule, explanatory notes, max_parallel_translations, and the scanned project, reusable by sas_translate() while sources are unchanged. ready_for_translation means no known missing inputs or unresolved scan findings; it does not mean execution or reference validation has passed.

Examples

source <- tempfile(fileext = ".sas")
writeLines("data out; set raw.dm; run;", source)
check <- sas_preflight(source, diagnose = "off")
check$inputs
check$model_calls
unlink(source)

Round the way SAS does

Description

SAS ROUND rounds half away from zero, to a rounding unit; R's round() is banker's rounding.

Usage

sas_round(x, unit = 1)

Arguments

Value

x rounded to the nearest multiple of unit.

See Also

Other runtime helpers: apply_format(), chr_cmp(), lib_delete(), lib_exists(), lib_members(), lib_read(), sas2r_fold_names(), sas2r_libname_assign(), sas2r_resolve_registry(), sas2r_source_include(), sas_compress(), sas_display(), sas_if_else(), sas_merge(), sas_min(), sas_missing(), sas_sort(), sas_sum()

Examples

sas_round(c(0.5, 1.5, -0.5))
sas_round(12.345, 0.01)

Sort a dataset the way PROC SORT does

Description

Stable, with missing values lowest under SAS collation; by is matched to column names case-insensitively, and variables named in descending are sorted in reverse.

Usage

sas_sort(df, by, descending = character(), ...)

Arguments

Value

df reordered.

See Also

Other runtime helpers: apply_format(), chr_cmp(), lib_delete(), lib_exists(), lib_members(), lib_read(), sas2r_fold_names(), sas2r_libname_assign(), sas2r_resolve_registry(), sas2r_source_include(), sas_compress(), sas_display(), sas_if_else(), sas_merge(), sas_min(), sas_missing(), sas_round(), sas_sum()

Examples

sas_sort(data.frame(id = c(2, NA, 1)), by = "id")
sas_sort(data.frame(id = c(2, NA, 1)), by = "id", descending = "id")

Row-wise SUM and MEAN with SAS missing semantics

Description

SAS SUM() and MEAN() are row-wise across their arguments: each observation gets the sum or mean of its non-missing arguments, and only an observation whose arguments are all missing yields missing. Column aggregation (⁠PROC MEANS⁠, ⁠PROC SQL⁠) is emitted inline by those emitters and never routed through these.

Usage

sas_sum(...)

sas_mean(...)

Arguments

Value

A numeric vector of that length.

See Also

Other runtime helpers: apply_format(), chr_cmp(), lib_delete(), lib_exists(), lib_members(), lib_read(), sas2r_fold_names(), sas2r_libname_assign(), sas2r_resolve_registry(), sas2r_source_include(), sas_compress(), sas_display(), sas_if_else(), sas_merge(), sas_min(), sas_missing(), sas_round(), sas_sort()

Examples

sas_sum(c(1, NA, 2), c(1, NA, NA))
sas_mean(c(1, NA, 3), c(3, NA, NA))

Translate SAS programs or project to R with dependency-aware orchestration

Description

Exposes the single authoritative SAS-to-R migration pipeline. SAS source is the only required input. Operates on a single SAS file (wrapped as a one-module project) or a directory of SAS files with identical internal sequencing: initializes/reconciles run -> discovers dependency graph and output contracts -> baseline generation and immediate program repair -> full bundle attempt execution and output-driven repair -> deterministic gate selection and reporting.

Usage

sas_translate(
  path,
  out_dir = NULL,
  config = NULL,
  execute = TRUE,
  max_program_repair_rounds = 1L,
  max_bundle_repair_rounds = NULL,
  outputs = NULL,
  agent_evidence = c("code_only", "bounded"),
  llm = NULL,
  budget_usd = Inf,
  budget_mode = "stop",
  pricing_source = "catalog",
  pricing_rates = NULL,
  usage_limits = NULL,
  recursive = FALSE,
  resume = FALSE,
  keep_raw_attempts = FALSE,
  max_bundle_repairs_per_component = 2L,
  max_parallel_translations = NULL
)

Arguments

Details

Missing inputs, called macros, includes and uncertain dependencies are reported while available source continues translating. Unavailable dependencies defer affected execution; they do not authorize inventing missing source or data. Component failures preserve useful work and allow other components to proceed. Unusable configuration, inconsistent pipeline coverage and critical provider or accounting failures still stop the run. Use sas_preflight() to inspect readiness and configure macros.search_path for external macro definitions.

Value

An object of S3 class "sas2r_translation" containing ⁠$run_id⁠, ⁠$out_dir⁠, ⁠$bundle_dir⁠, ⁠$outputs_dir⁠, ⁠$status⁠, ⁠$status_reason⁠, ⁠$graph_path⁠, ⁠$output_contracts_path⁠, ⁠$report_path⁠, ⁠$report_json_path⁠, ⁠$component_evidence⁠, ⁠$output_assessments⁠, ⁠$diagnostics⁠, ⁠$repair_history⁠, ⁠$usage⁠, and ⁠$project⁠. ⁠$outputs_dir⁠ contains selected contract deliverables with their library and relative-path layout (for example datasets/work/out.rds, datasets/adam/adsl.rds, and tlf/outputs/table.html); it is NULL when execution is disabled or no selected deliverables were produced. The JSON report includes separate target/reference/review coverage and effective limits.

Examples

# Translate a small SAS program with the deterministic rule-based engine.
# Neither an LLM nor a SAS installation is required.
sas_dir <- file.path(tempdir(), "sas2r-translate-example")
dir.create(sas_dir, showWarnings = FALSE)
writeLines(c(
  "data work.adsl;",
  "  set adam.dm;",
  "  where age >= 18;",
  "  bmi = weight / (height * height);",
  "run;",
  "",
  "proc sort data=work.adsl out=work.adsl_sorted;",
  "  by usubjid;",
  "run;"
), file.path(sas_dir, "adsl.sas"))

# execute = FALSE translates without running the generated bundle.
res <- sas_translate(sas_dir, out_dir = tempfile("sas2r-out"), execute = FALSE)
res$status
cat(sas_code(res))

## Not run: 
# Run the generated bundle and let the agents translate, review, and repair
# the patterns the deterministic rules cannot prove.
llm <- sas_llm(list(provider = "anthropic", model = "claude-sonnet-4-6"))
res <- sas_translate("path/to/sas/project", llm = llm, recursive = TRUE)
sas_write(res, "path/to/r/bundle")

## End(Not run)

Write translated code artifacts and outputs to a destination directory

Description

Copies the selected editable bundle (including failed code), with its ⁠programs/⁠, ⁠macros/⁠, ⁠runtime/⁠, autoexec.R, run.R, and README. Saved automated deliverables go to ⁠saved-outputs/⁠ and reports to ⁠report/⁠. Existing manual ⁠output/⁠ files are not copied; configuration edits are kept. Input libraries remain external dependencies documented in the README. Run ⁠Rscript run.R⁠ from the exported folder, or source("run.R", chdir = TRUE). Warns if status is blocked or needs_review.

Usage

sas_write(x, dir, overwrite = FALSE)

Arguments

Value

The target directory path, invisibly.

Examples

sas_dir <- file.path(tempdir(), "sas2r-write-example")
dir.create(sas_dir, showWarnings = FALSE)
writeLines(c("proc sort data=work.dm out=work.dm_sorted;",
             "  by usubjid;",
             "run;"), file.path(sas_dir, "sort.sas"))
res <- sas_translate(sas_dir, out_dir = tempfile("sas2r-out"), execute = FALSE)

# Writing an unverified translation warns; the bundle is still written.
dest <- tempfile("sas2r-bundle")
suppressWarnings(sas_write(res, dest))
list.files(dest, recursive = TRUE)

Split a SAS dataset name into its library and member

Description

This internal helper is available in a generated bundle's runtime. Use the returned names exactly; it returns a character vector, not a list. Accessing an absent name with [[ raises an error, rather than returning NULL.

Usage

split_ds(ds, macro_vars = character())

Arguments

Value

A named character vector of length two, with elements lib and member. A one-level name uses lib = "work". There is no libref element.

Write a PROC-COMPARE-style markdown report or versioned JSON comparison report

Description

Write a PROC-COMPARE-style markdown report or versioned JSON comparison report

Usage

write_comparison_report(
  x,
  file = NULL,
  base_label = "SAS (base)",
  comp_label = "R (compare)",
  run_id = "default",
  dir = NULL,
  ...
)

Arguments

Value

Path to the written file, invisibly.

Examples

sas <- data.frame(usubjid = c("01-001", "01-002"), aval = c(1.0, 2.0))
r <- data.frame(usubjid = c("01-001", "01-002"), aval = c(1.0, 2.5))
cmp <- compare_datasets(sas, r, keys = "usubjid")

# PROC COMPARE-style markdown.
md <- file.path(tempdir(), "sas2r-comparison.md")
write_comparison_report(cmp, file = md)
cat(head(readLines(md), 6), sep = "\n")