SAS static analysis

The SAS parser is separate from SQLGlot. It groups semicolon-delimited DATA/PROC blocks, preserves line locations, strips comments safely, resolves configured/static %let values, and reports unsupported runtime constructs. PROC SQL statements use SQLGlot after SAS handling.

Tested: DATA outputs, SET/MERGE/UPDATE/MODIFY inputs, explicit INFILE/FILE paths, WORK assets, PROC SQL CTAS, SORT OUT= (or in-place), SUMMARY/MEANS OUTPUT OUT=, TRANSPOSE, IMPORT DATAFILE/OUT, EXPORT DATA/OUTFILE, common DATA/OUT procedure conventions and APPEND DATA/BASE. PROC SQL SELECT expressions accept literal FORMAT= display modifiers (named, numeric, character and custom formats ending in a period or decimal precision), before or after aliases, and positive literal LENGTH= modifiers. CALCULATED references use uniquely defined SELECT aliases without inventing base-table columns; missing or ambiguous aliases produce diagnostics. SAS DATETIME() is recognized as an expression. Malformed modifiers remain parse errors. These rules customize only the configured PROC SQL dialect instance; ordinary SQL is unchanged. See the SAS column modifier reference.

Parser cache version 11 invalidates earlier results on the next analysis. Run Analyze all files in Analysis for previously analyzed workspaces; source bytes and original statement/line evidence remain intact. Existing packaged executables/images need rebuilding for version 11.

Column transformation subset

Transformation lineage reuses SQL derivations for PROC SQL and adds a bounded DATA-step subset: explicit assignments with ordered variable versions, single SET propagation and explicit KEEP/DROP/RENAME when statically established. Original assignment and PROC SQL expression text remains cited. Unknown passthrough schemas remain visible; a partial list is not promoted to a complete schema. MERGE BY, multiple SETs, arrays, loops, conditional retention, automatic variables and runtime/macro-expanded expressions are unresolved. Existing table-level MERGE recognition does not imply SQL join semantics or a compiled SAS variable analysis. No SAS code executes.

Static macro-variable expressions

%LET can combine already-known direct references with literal text, for example %let lib=raw; %let source=&lib..orders;. Values resolve when assigned, so changing lib later does not change source. References in double-quoted paths expand; single-quoted paths remain literal. Original DATA/PROC code and line ranges retain the macro references. See SAS's macro-variable reference rules.

Unknown/indirect references, macro functions and oversized expansions remain unresolved. An unresolved reassignment invalidates the previous variable value. Unmodeled macro calls, conditional macro flow and CALL SYMPUT/SYMPUTX/EXECUTE invalidate known variable bindings conservatively; a later explicit %LET can establish a fresh value. Configured values cannot inject additional semicolon-delimited statements. Expansion is bounded to 1 MiB per statement. This adds static variable expressions, not macro-definition invocation or a SAS interpreter.

Library mappings rewrite namespaces, not filesystem permissions. LIBNAME paths are not read. One-level SAS datasets use WORK; WORK objects are temporary but remain in internal lineage.

Direct variables are conservative explicit KEEP/DROP/BY/RENAME lists and procedure VAR/ID/BY where a single source establishes ownership. Multiple-source unqualified variable ownership is not guessed. SAS ranges, arrays, automatic variable lists and arbitrary expressions do not constitute a full compiled variable analysis. Consult source evidence for these constructs.

Macro definitions/invocations and conditional macro blocks are retained and diagnosed without execution/expansion. Unresolved macro inputs suppress uncertain block edges. Name literals outside the plain identifier rules are unresolved instead of tokenized into fabricated tables. DATALINES/CARDS (including 4-semicolon forms) are data payloads and are never parsed as code.

Limitations: no SAS runtime, macro interpreter, PROC IML, general procedure output semantics, dynamic libraries, automatic runtime session discovery or faithful execution order. Static WORK datasets use workspace and explicitly configured job/folder scope. Configure supplied job IDs in Parsing and lineage or use separate workspaces for unrelated sessions; absent that context, independent jobs can remain ambiguous. Optional AI does not bypass evidence validation.

DATA/PROC evidence is sliced directly from the original source between statement offsets; spacing, blank lines, comments and same-line block boundaries remain intact. Inline-data contents remain evidence text without becoming executable statements or guessed dependencies.