mpfr_expr_cli tutorial

This tutorial shows how to check bin_float against an MPFR data file from the command line. The runner is reached as floating-conformance --backend mpfr; below, mpfr abbreviates that command.

Quick start

just conformance build binary
_build/conformance/mpfr/native/release/build/mpfr-conformance.exe \
  --backend mpfr testdata/bin_float/mpfr-4.2.2-elementary.txt

MPFR elementary summary
cases: …
passed cases: …
failed cases: 0

Everyday tasks

Run each format

The runner reads exactly one file and picks the format from its content:

mpfr testdata/bin_float/mpfr-4.2.2-elementary.txt   # contains "mpfr-elementary-v1"
mpfr pow_si.txt                                     # header contains "input_coefficient_hex"
mpfr path/to/mpfr/tests/data/sqrt                   # anything else: sqrt data_check

JSON for scripts

mpfr --json testdata/bin_float/mpfr-4.2.2-elementary.txt

prints one object with corpus, totalCases, passedCases, failedCases and failedIds.

Going further

just conformance run binary runs the pinned MPFR square-root data and the elementary matrix together with the TestFloat matrix.
New elementary data is produced by tools/generate_mpfr_elementary_oracle.c; keep the mpfr-elementary-v1 marker line so the runner detects the format.
mpfr_expr tutorial explains the row formats and pass rules.

Common pitfalls

One file per invocation. Extra paths, missing paths, and the shard options all print the usage line and exit with 2.
Format detection is textual. A square-root file that happens to contain input_coefficient_hex in a comment is parsed as power data.