"""Single entry point for the `silurian` study: regenerates every number and figure.

    python3 run_all.py              # download anything missing (cached), then run every step
    python3 run_all.py --no-fetch   # use the raw cache only (fails if a raw file is missing)
    python3 run_all.py s04 s07 s08  # run selected steps only

Raw downloads go to $SILURIAN_RAW (default fermi_raw/silurian); small derived
tables go to ../data, figures to ../figures.  A cold start downloads about 210 MB and needs
roughly 20-40 minutes, almost all of it the 4,000 Macrostrat map-point queries and the PBDB
collection lists; with a warm cache the whole pipeline takes about three to five minutes on one core.

Steps
  s00_fetch       Macrostrat columns/units/sections + 4,000 random map points, PBDB collections,
                  CENOGRID (PANGAEA 917660), DM-SED (Zenodo 14771859), Natural Earth land mask
  s01_macrostrat  preserved sediment vs age, hiatuses, yr/cm (time-weighted and plain medians),
                  Sadler slopes
  s02_mapsample   global bedrock-map point sample: outcrop by age, North America -> globe scaling
  s03_pbdb        fossil-collection density; space-time probe P(sampled | footprint, window)
  s04_cenogrid    injection-recovery of industrial carbon pulses with two detectors (matched filter
                  and a plain step statistic); sample-spacing coverage of the record itself
  s05_dmsed       density and spacing of geochemical sampling in one deep-time compilation
                  (exact interval union; lower bounds)
  s06_signals     signal magnitudes: carbon, radionuclide windows, conditional 244Pu bound with the
                  modern-layer test, lunar erasure
  s07_synthesis   p(known today) by scenario, age and window, split into an assumed surface part and
                  a data-based deep-time part; marginalised exclusion counts; modelling variants
  s08_figures     fig1-fig5

Revision history: the first pass was reviewed by a referee (../referee_report.md); the changes made
in response are listed in the module docstrings of s01, s04, s05, s06 and s07 and in the report.

Random seeds are fixed (common.SEED), so a rerun on the same raw files reproduces the tables
exactly.  PBDB and Macrostrat are living databases: a fresh download at a later date will
shift counts by a small amount.
"""
import os, sys, time, subprocess

HERE = os.path.dirname(os.path.abspath(__file__))
STEPS = ["s00_fetch", "s01_macrostrat", "s02_mapsample", "s03_pbdb", "s04_cenogrid", "s05_dmsed",
         "s06_signals", "s07_synthesis", "s08_figures"]


def main(argv):
    args = [a for a in argv if not a.startswith("--")]
    steps = [s for s in STEPS if not args or any(s.startswith(a) for a in args)]
    if "--no-fetch" in argv:
        steps = [s for s in steps if s != "s00_fetch"]
    logdir = os.path.join(os.path.dirname(HERE), "data")
    os.makedirs(logdir, exist_ok=True)
    t00 = time.time()
    with open(os.path.join(logdir, "run_all.log"), "w") as log:
        for s in steps:
            t0 = time.time()
            print(f"[run_all] {s} ...", flush=True)
            p = subprocess.run([sys.executable, os.path.join(HERE, s + ".py")], cwd=HERE,
                               stdout=subprocess.PIPE, stderr=subprocess.STDOUT, text=True)
            log.write(f"===== {s} (exit {p.returncode}, {time.time() - t0:.0f} s) =====\n{p.stdout}\n")
            log.flush()
            if p.returncode != 0:
                print(p.stdout[-4000:])
                raise SystemExit(f"[run_all] {s} failed (exit {p.returncode})")
            print(f"[run_all] {s} done in {time.time() - t0:.0f} s", flush=True)
    print(f"[run_all] all steps finished in {time.time() - t00:.0f} s; log: {os.path.join(logdir, 'run_all.log')}")


if __name__ == "__main__":
    main(sys.argv[1:])