FLOOXS » TclLib » Verify
13 documented proc(s) in TclLib/Verify/.
::Verify::finite · ::Verify::fmt · ::Verify::peak_error_field · ::Verify::record · verify_assert · verify_close · verify_count · verify_failures · verify_newton_residual · verify_newton_update · verify_relative · verify_report · verify_reset
::Verify::finite — is `v` a real, finite number? Guards the comparison arithmetic: Tcl's abs() raises on NaN, and a NaN or Inf measured value must be recorded as a FAILED check rather than aborting the deck with a domain error. Returns 1 for a finite double, 0 for NaN / +-Inf / anything non-numeric.
::Verify::fmt — 6-significant-digit rendering for the report lines. Non-numeric values (a NaN that arrived as a string, an empty field read) pass through verbatim so the log shows what actually happened.
::Verify::peak_error_field — max |value| of <Sol>_<suffix> over `mater`. The two nodal error fields FLOOXS writes at every Newton update (src/BasePDE/Genpde.cc:868-885 create them, :1002-1003 fill them) are named after the SOLUTION, not the material, so one field per solution covers every mesh. Rows the contact BC has remapped are skipped by the update loop (Genpde.cc:914) and read back as 0, so the peak is over interior nodes only — which is exactly the set a convergence statement is about. The absolute value is taken inside `sel` so `peak` spends its digits on the magnitude rather than on a sign; a missing field is a hard error rather than a silent 0, because a bound derived from a field that was never written would be meaningless.
::Verify::record — the one place a check enters the log. `measured` is the quantity being bounded (already reduced to a single number by the caller), `bound` its allowed magnitude, `just` the required derivation of that bound. Prints the structured PASS/FAIL line immediately so the ordering in the test log matches the deck, and returns 1 on pass / 0 on fail.
verify_assert — assert |measured| <= bound. The primitive. `measured` is a residual, imbalance, or error the deck has already reduced to one number; `bound` is its derived allowance; `justification` states where that allowance comes from and is REQUIRED (blank -> TCL_ERROR). Returns 1 on pass, 0 on fail.
verify_close — assert |measured - expected| <= bound. Absolute-difference form for comparing against an analytic or independently-computed value. Same justification requirement.
verify_count — number of checks recorded so far (passed and failed).
verify_failures — number of recorded checks that failed so far.
verify_newton_residual — largest converged-state equation residual. Same field family, the right-hand-side half: <Sol>_NewtonRHSError holds |r_i| for the last assembled residual, in the units of the assembled ROW. Good for a convergence statement about the equation itself, and for showing that a re-solve does not grow the residual. NOT usable, today, to bound a terminal CURRENT: the relationship between a bulk row's units and the `contact … flux` readback (Solver::ComputeBoundaryReaction multiplies the flux row by FluxScale × Bound::Scale) does not reproduce the measured terminal imbalance — see Test/Verify/README.md, "Nodal error fields".
verify_newton_update — largest converged-state Newton update, in the solutions' own units. THE POINT. A bound derived from `math updateLimit` and the per-solution abserr/relerr states the largest update the convergence TEST would have accepted. The solve almost always does far better than it is asked to, and the solver records by how much: this reads the update actually applied on the final Newton step out of <Sol>_NewtonUpdError. Using it in place of the a-priori criterion typically tightens a converged-state bound by three to four orders while keeping the same derivation chain. CAVEAT, and it is the reason this is a bound and not an identity: the stored value is the update that PRODUCED the reported state, i.e. the distance from the previous iterate. Near a simple root Newton contracts, so the remaining distance to the exact discrete solution is no larger — conservative, but it is an assumption about the local convergence rate and belongs in the deck's justification string. set du [verify_newton_update Silicon {DevPsi Qfn Qfp}] Read it immediately after the solve whose state is being measured; a cached bias step (`device_cache` with `norunonexists`) restores a structure WITHOUT solving, leaving the field describing an older state.
verify_relative — assert |measured - expected| / |expected| <= bound. Relative form, for quantities whose scale is set by the physics rather than by the discretization. `expected` == 0 is a usage error (the relative error is undefined there) and raises TCL_ERROR — use verify_assert with an absolute bound for near-zero targets.
verify_report — print the summary and the harness verdict. Re-lists every failed check (so the failures are together at the end of a long log), prints the `Passed` / `Failed` line the CTest PASS_REGULAR_EXPRESSION looks for, and exits 1 when anything failed. A deck that never calls this prints no `Passed` line and therefore fails CTest — that is deliberate.
verify_reset — clear the recorded checks and failure count. Call once at the top of a deck. Also used by decks that deliberately exercise the failure path (see Test/Verify/verify_selftest.tcl) to drop the intentional failures before the real report.