__file__ and sys.executable from the resolved executable path

• Type: feature (frontend / runtime) — Track N
• Opened: 2026-08-13, the implementation half of [[decide-nilpy-dunder-file-for-a-compiled-program]], decided by the user the same day. Read that ticket first: it records why the compile-time source path was rejected (it leaks the build environment into every shipped binary) and why pxx follows the FREEZER convention (PyInstaller/cx_Freeze) rather than CPython's source convention.

What to build

1. Resolve the executable, do not trust argv[0]. argv[0] can be a PATH lookup, a relative path, or whatever an exec caller chose to pass. /proc/self/exe on hosted Linux; resolve argv[0] against PATH/CWD on any target without it. One helper, used by both names below.
2. __file__, main module = that resolved executable path. os.path.exists(__file__) is then True, which is worth having.
3. __file__, imported module = <exe_dir>/<original module basename> — e.g. /opt/app/labels.py. A VIRTUAL path: no file is there. Deterministic, leaks nothing, and makes os.path.dirname(os.path.abspath(__file__)) — the only form that matters in practice — the executable's directory for every module.
4. sys.executable = the same resolved path. Small, and it takes "where am I installed" off __file__ for good.

Today __file__ is argv[0] unresolved, and sys.executable does not exist.

Explicitly NOT in scope

--data-root / a run-time base-directory override. Decided (user) to wait for the first program that actually needs it rather than building it speculatively; the decision ticket records the shape so it can be added without re-litigating.

Divergence, already documented

open(__file__) on an imported module fails — there is no file at that path. Frozen Python behaves the same once its temp dir is gone. The entry is already in devdocs/dev/nilpy-semantics-divergences.md ("__file__ names the EXECUTABLE, not the source"); check it still matches what you land, and fix the doc if it does not.

The user-facing half is [[docs-nilpy-file-dunder-and-data-files]] (Track D), blocked on this ticket. Unblock it when this lands.

Gate

A .npy that prints __file__, dirname(abspath(__file__)) and sys.executable for the main module AND for an imported sibling module, run from a different directory than the one it was compiled in — that last part is the whole point, and it is what the existing corpus habit hides (see the decision ticket's note on why this took until 2026-08-13 to surface). Plus test-nilpy green and the self-host fixedpoint.

Worth checking in the same pass: any other place the frontend answers a "where am I" question from an unresolved argv[0].

Log

• 2026-08-14 — resolved, commit 5a71fe0be.