Abort when post_process is given a different grid than the restart files

[sbryngelson]

post_process checks that the restart files exist but never that they hold the grid the case file asks for. Give it a case whose resolution no longer matches the run and it reads past the end of every file, post-processes the overrun, and exits 0 with NaN-filled output. Nothing reports a problem until someone plots it — or until a diagnostic built on that output starts returning NaN and gets blamed instead, which is how I found it.

Reproduction

A 349 M cell run (m = 1644) post-processed with a case file that had since moved to m = 2056:

simulation.inp:    m = 1644     <- wrote the restart files
post_process.inp:  m = 2056     <- read them

681 M cells read from files holding 349 M. Exit code 0, silo written, every field NaN.

The check

x_cb.dat holds exactly m_glb + 2 cell boundaries, so its size states which grid wrote the restart. One inquire on a file already being opened, before any bulk read:

Restart grid mismatch: this case has m = 2056 but
.../restart_data/x_cb.dat was written with m = 1644. Post-processing must use the
same grid as the run that wrote the restart files, or it reads past the end of
every file and writes NaN.

Verified against the failing data: 2058 boundaries would need 16464 bytes, the file holds 13168, so it aborts where it previously ran to completion.

Builds clean (./mfc.sh build --no-gpu -t post_process).

https://claude.ai/code/session_01HMJ7cycfo7kTFSFq5yhHLG

[Copilot]

Warning

Copilot couldn't run its full agentic review because it didn't start before the timeout. Make sure your repository has a runner available, or add a copilot-code-review.yml file specifying one with the runs-on attribute. See the docs for more details.

Pull request overview

Adds safety checks and new runtime features around restart handling, boundary-condition ramps, and immersed-boundary (IB) output/kinematics.

Changes:

• Abort post-processing early when restart grid resolution (from x_cb.dat size) doesn’t match the case file’s m.
• Add ramped inflow support (vel_in_ramp, vel_in_t0, vel_in_frac0) for GRCBC and Dirichlet-style inflows.
• Add prescribed IB kinematics options and buffered per-step IB force/kinematic logging (D/ib<id>_forces.dat) with ib_force_stride.

Reviewed changes

Copilot reviewed 16 out of 16 changed files in this pull request and generated 11 comments.

Show a summary per file

File	Description
toolchain/mfc/params/descriptions.py	Adds parameter descriptions for IB force stride and IB kinematics.
toolchain/mfc/params/definitions.py	Registers new params/hints (inflow ramp, IB force stride, IB kinematics) and raises probe max.
toolchain/mfc/case_validator.py	Validates inflow ramp constraints and IB stride/kinematics constraints.
src/simulation/m_time_steppers.fpp	Calls inflow-ramp updater and IB force logging; integrates prescribed kinematics into RK stages.
src/simulation/m_start_up.fpp	Flushes buffered IB force records on finalize.
src/simulation/m_mpi_proxy.fpp	Broadcasts new BC ramp parameters and IB kinematics parameters.
src/simulation/m_ibm.fpp	Initializes prescribed kinematics at start; adds kinematics implementation and adjusts GPU/MPI data movement.
src/simulation/m_global_parameters.fpp	Initializes new global/BC/IB parameters (ib_force_stride, inflow ramp fields, kinematics fields).
src/simulation/m_data_output.fpp	Implements buffered per-step IB force/kinematics output and adjusts probe output formatting.
src/simulation/m_cbc.fpp	Stores “final” inflow velocity and applies time ramp factor to inflow + exposes updater.
src/pre_process/m_global_parameters.fpp	Initializes new IB kinematics fields for preprocessing.
src/post_process/m_data_input.f90	Adds restart grid mismatch detection via x_cb.dat file size.
src/common/m_derived_types.fpp	Extends BC and IB derived types with inflow ramp + kinematics parameters.
src/common/m_constants.fpp	Raises num_probes_max to 64.
src/common/m_boundary_primitives.fpp	Applies ramp factor to Dirichlet boundary ghost velocities via bc_vel_ramp.
docs/documentation/case.md	Documents new IB kinematics, IB force logging, and inflow ramp parameters.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[sbryngelson]

Rebased onto master as one commit. The branch had been cut from a working branch and was dragging along stale, divergent copies of #1850, #1846, #1847 and the probe changes — including a bc_vel_ramp module array with GPU_DECLARE(create=...) and no device update, which under OpenACC would multiply ghost-cell velocities by uninitialized device memory on any Dirichlet boundary. That code does not exist in #1850, which evaluates the ramp on device from mytime through a pure function; it was an older draft that should never have been on this branch. Gone now — the PR is the 19-line check and nothing else.

Two real bugs in the check itself, both fixed:

Hardcoded 8 bytes. x_cb.dat is written with mpi_p, and mpi_p follows wp, so a --single build writes four-byte reals and every single-precision post_process run would have aborted, reporting half the true m. The --single CI lane runs --test-all, which includes post_process, so this would have failed CI. Now taken from storage_size(0._wp)/8.

down_sample broke. m_start_up.fpp:83 reduces m to int((m+1)/3) - 1 before m_glb = m, while x_cb.dat still holds full-resolution boundaries, so the strict equality fired spuriously. examples/3D_IGR_33jet/case.py sets it and no test covers it. The read is strided by 3 and touches stride*(m_glb + 1) + 1 boundaries, so down-sampling needs more of the file, not less; the check now compares against that, and keeps the exact equality only for the un-strided read — three source grids of different size can down-sample to the same m_glb, so exact is not available there.

./mfc.sh build -t post_process is clean. Dropped the Claude-Session trailer.

[github-actions]

Claude Code Review

Head SHA: 4007be2

Files changed:

• 16
• docs/documentation/case.md
• src/common/m_boundary_primitives.fpp
• src/common/m_constants.fpp
• src/common/m_derived_types.fpp
• src/post_process/m_data_input.f90
• src/pre_process/m_global_parameters.fpp
• src/simulation/m_cbc.fpp
• src/simulation/m_data_output.fpp
• src/simulation/m_global_parameters.fpp
• src/simulation/m_ibm.fpp
  (+5 more)

Findings:

• src/post_process/m_data_input.f90: the new restart-grid size check hardcodes 8 bytes per real (int(m_glb + 2, 8)*8_8), but the grid file is read via mpi_p/real(wp) whose width depends on the build's working precision (m_precision_select). In a single-precision build, x_cb.dat holds 4-byte reals, so file_bytes will never equal (m_glb+2)*8, and this check will abort every valid single-precision restart with a spurious "grid mismatch" error.
• src/common/m_boundary_primitives.fpp: the new Dirichlet ramp scaling only touches momentum components with bc_vel_ramp(1) in the single added loop (hardcoded index 1, not a per-direction variable), even though bc_vel_ramp is declared dimension(3) and the feature (bc_[x,y,z]%vel_in_ramp, validated and documented for x, y and z alike) is meant to apply per direction. Unless equivalent scaling is also wired up at the y/z Dirichlet ghost-cell fill sites (not present in this diff), a bc_y%vel_in_ramp/bc_z%vel_in_ramp ramp on a Dirichlet boundary will be silently ignored while bc_x%vel_in_ramp works.
• src/simulation/m_data_output.fpp (s_write_ib_force_files): the record's body id is computed as real(max(patch_ib(ib_idx)%gbl_patch_id, ib_idx), wp), mixing a local array index (ib_idx, from local_ib_patch_ids(i) in multi-process runs) with the global id via max(). If a rank's local index for a body exceeds that body's true gbl_patch_id (a realistic case once bodies are distributed/reordered across ranks), max silently picks the local index instead of the global id, so the row is written to the wrong D/ib<id>_forces.dat file.

[github-actions]

Lines of Code

File	Lines	Diff
src/post_process/m_data_input.f90	431	+11

Directory	Lines	Diff
post_process	3399	+11
total	46351	+11

[codecov]

Codecov Report

❌ Patch coverage is 33.33333% with 4 lines in your changes missing coverage. Please review.
✅ Project coverage is 61.25%. Comparing base (dc0aec1) to head (6f4294c).

Files with missing lines	Patch %	Lines
src/post_process/m_data_input.f90	33.33%	3 Missing and 1 partial ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #1858      +/-   ##
==========================================
- Coverage   61.26%   61.25%   -0.01%     
==========================================
  Files          84       84              
  Lines       22330    22335       +5     
  Branches     3265     3266       +1     
==========================================
+ Hits        13680    13681       +1     
- Misses       6207     6210       +3     
- Partials     2443     2444       +1     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.