Conversation
- Exclude -R from the rebuilt legacy argv[] (already consumed by GMT_Parse_Common into GMT->common.R): fixes "Unknown option -R..." - resamplegrid()/GMT_get_bcr_z() now skip land source nodes in the bicubic stencil, renormalizing over wet nodes only. Coastal terrain elevations stored in etad on dry nodes were bleeding into nearby wet-node interpolates as invented negative sea-surface values, triggered only via -J's (now -P's) nested-grid hold/resync path. - Rename time-jump option -J to -P: GMT_Create_Options() unconditionally prefixes '+' to any -J argument starting with a digit (EPSG-code shorthand support in gmt_parse.c), which mangled a numeric -J<time_jump> into -J+<time_jump> before nswing's own parser ever saw it, making run_jump_time silently pick up time_jump's value. - Remove the old -O<int>,<outfname> (unused maregraph-interval) option; rename -B<BCfile> to -O<BCfile>. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Replace the MEX-era argv[] rebuild + switch(argv[i][1]) dispatch with the standard GMT pattern: iterate options directly and switch(opt->option), reading opt->arg. Drops parse()'s argc_out/argv_out signature (now matches other GMT modules' parse(GMT, Ctrl, options)). -Fk's region decode no longer mutates opt->arg in place with a strdup/strcat restore hack; uses a small local scratch buffer instead. The netCDF History attribute is now built via GMT_Create_Cmd()/ GMT_Destroy_Cmd() instead of hand-joining a rebuilt argv[]. Verified: clean build, basic run + -Fk region/RxC parsing produce identical output/History attribute to before. A crash in the multi-prism (-Fk.../RxC) simulation loop was found but confirmed pre-existing (reproduces on the prior commit too, unrelated to this change). Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
-Fk<region>/RxC or /dx/dy forces out_maregs_nc since the whole point of a prism grid is sampling a maregraph per prism, but nothing checked that -T (which sizes/allocates the maregraph arrays via n_mareg/n_ptmar) was actually given. Without it those stayed at their uninitialized stack values, calloc'd a garbage-sized buffer, and corrupted memory once real maregraph data got written into it (previously misread as a crash in the kaba goto-loop itself; bisected against the prior commit and confirmed it wasn't the parse() refactor). Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
nswing copies every grid into its own flat arrays and handles boundary conditions itself; GMT's default grid pad is dead weight here, and the user wants it gone outright rather than relying on gmt_M_ijp() to always index around it. GMT_Set_Default(API, "API_PAD", "0") applied once right after GMT_Parse_Common, before any grid is read or created. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The parse() rewrite to switch(opt->option) dropped the -R exclusion that used to live in the old argv-rebuild loop, so -R fell through to default: and errored as "Unknown option". Restored as a continue filter before the switch (not a case label) -R is already consumed by GMT_Parse_Common into GMT->common.R and isn't one of nswing's own options. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
When no bathy grid is given but a source (-F or -Fk) and -R are, there's nothing to simulate -- skip the whole time-stepping machinery and just compute the deformation directly over -R's grid geometry (region + increment + registration, populated by GMT_Parse_Common when -R names a grid file) and write it to -G's name. -t and -G's saving interval are no longer required in this case. Verified: `-Rbathy.grd -F<9 params> -Glixa_def.grd` now writes a grid matching bathy.grd's exact geometry with sane deformation values, no simulation run; the normal simulation path (bathy given) still requires -t as before. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The big one: mass() had 'for (row = row_start; row_end; row++)' -- the loop condition was the constant row_end, so the row loop ran past the array end and every simulation segfaulted in the first mass_conservation() call since the Mirone conversion (2380d50). Only the deformation-only mode worked. Also fixed, from a cppcheck + manual audit: - -Em always aborted: '||/&&' precedence bug skipped the workMax calloc and fell into the out-of-memory branch. - work array sized with a comma expression '(a, b)' instead of MAX(a, b). - sanitize_nestContainer typo left fluxn_d[] uninitialized (fluxm_d was NULLed twice). - '!stem' (array, always false) dead check replaced by a scoped test; guarded the strlen(stem)-1 underflow when stem is empty. - always-true '(nest.long_beach || nest.short_beach)' array-decay test replaced by the do_long_beach/do_short_beach flags. - initialize_nestum() read nest->hdr[lev-1] before the lev > 0 guard. - -X Manning parsing rewritten with strtok, bounded to the 10 levels the struct holds (was an unbounded manning[k++] overflow, plus a hidden t[16] overflow). - parse() hardening: field widths on all sscanf %s, bounded strncpy for -S, bounds guard on the -G '.nc' extension append, short-arg guard in -L, Okada params zero-initialized, deformation-only mode now requires -G explicitly. - error-path cleanups (matter when called repeatedly from a persistent GMT.jl session): destroy read grids on the error bail, only open the ASCII maregraph file when not writing netCDF, free mareg_names and the tracers container on their failure paths, close fp_oranges, call free_arrays if initialize_nestum fails, guard the netCDF maregs end block when no maregraphs were found, require a bathy grid of at least 2 x 2 nodes. - check_paternity(): loop-safety only (k++ and k < 10 bound). The check itself stays inert: reviving it rejected valid nested grids because check_binning()'s math predates the current half-cell nesting rules. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Replace struct srf_header (a relic of the old Surfer-binary I/O) with the generic struct grd_header everywhere: gmtnswing_get_grid(), deform(), kaba_source(), the hdr_b/hdr_f/hdr_mM/hdr_mN headers and the nested-grid read block. Grids read through the API are no longer capped at 32767 rows/columns by the short int nx/ny fields, and the (short) casts in deformation-only mode are gone. gmtnswing_get_grid() now also fills nm and x_inc/y_inc straight from the GMT header, and the memset in kaba_source() computes its size in size_t. Deformation output verified bit-identical to the previous build; plain simulation and a real 2-level nested run pass. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…rflow Drop the home-grown struct grd_header and use GMT's own GMT_GRID_HEADER throughout (374 field references: nx/ny -> n_columns/n_rows, x_min etc -> wesn[], x_inc/y_inc -> inc[], plus the ij_grd macro). gmtnswing_get_grid() now just struct-copies *G->header. The write-only hdr.lat_min4Coriolis copies are gone (the nestContainer keeps its own). Also fix a long-standing heap overflow found while validating this: open_most_nc() allocated basename with malloc(strlen(base)) - no room for the NUL terminator, and the MOST variants strcat a further "_ha.nc" type suffix onto it. Every 3D/MOST netCDF run wrote past the block; the header refactor reshuffled the heap enough for the CRT to start aborting at teardown on some -G name lengths. Now strlen(base) + 8. Validation: single-threaded old vs new outputs are bit-identical (the default multithreaded runs cannot be compared - moment_M/moment_N race makes results nondeterministic run-to-run, a separate pre-existing issue). Deformation-only output bit-identical; nested 2-level case passes 8/8. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
-T<x/y>[+o<outmaregs>][+t<int>] is now equivalent to giving a one-point positions file: the token before the modifiers is taken as coordinates when it parses as exactly two /-separated numbers, otherwise it is a file name as before. The node index is computed with the same irint((x - wesn)/inc) math and inside-grid check that read_maregs() uses. Useful for quick single-gauge runs without touching disk. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The #ifndef PARALLEL prologue of the cartesian moment_N() did memset(fluxm_d, ...) - moment_M's output array - instead of its own fluxn_d. Under DO_MULTI_THREAD moment_M and moment_N run concurrently, so the wipe landed at a random point while moment_M was filling fluxm_d: cartesian results were nondeterministic run to run (up to ~0.27 m differences at a gauge). In a serial build the wipe simply zeroed the x-fluxes each step. The three sibling functions (moment_M, moment_sp_M, moment_sp_N) and the PARALLEL-mode hoisted memsets all use the right array, which is why geographic (-fg) runs were always deterministic and unaffected. Verified after the one-word fix: identical cartesian commands are now bit-identical run to run, a serial build matches the threaded one, and geographic output is unchanged. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Drop both the DO_MULTI_THREAD path (2 Windows threads created and destroyed every time step, moment only) and the never-enabled PARALLEL path (per-step slice threads whose integer-division row splits silently skipped up to n_threads-1 rows). In their place, OpenMP parallel-for with schedule(dynamic, 8) on the six row loops: mass, mass_sp, moment_M, moment_N, moment_sp_M and moment_sp_N. mass is now parallelized too, all cores participate, and threads persist in the OpenMP pool instead of being respawned each step. About 330 lines of scaffolding removed: MT_* wrappers, *_slice function variants and their mid-function #ifdef body-sharing tricks, ThreadArg, the PFV function pointers. -x<n> now maps to omp_set_num_threads() [default all cores]. dynamic scheduling matters on hybrid P/E-core CPUs, where static splits made every per-step barrier wait for the slowest efficiency core and ran slower than the old 2-thread code. Results are bit-identical to the old build and across schedules and thread counts (the loops have no reductions), so runs are finally reproducible regardless of -x. The four moment loops also zero dd/df per cell now: the goto L121/L201 shortcuts could read them stale from the previous cell (a latent NaN source for -S velocities, and a loop-carried dependency the parallel loops must not have). Timings (20-core machine, maregraphs-only runs): 765x476 geographic, 1010 steps: 5.8-7.3 s old -> 3.5 s; 1201x1201 cartesian, 300 steps: 12.0 s old -> 4.4 s. Non-Windows builds gain parallelism for the first time (old threading was Windows-only). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Two-stage detection, applied in both the simulation path and the deformation-only mode: first trust GMT's own grid metadata (gmt_M_is_geographic, set by the netCDF reader when the grid carries degree units), then fall back to the old -180/360, -90/90 range heuristic for unit-less grids (e.g. Mirone-written ones). Either way a warning is issued and isGeog is set. The deformation-only mode previously had NO detection at all - it returns before the bathymetry-reading branch where the heuristic lived, so 'nswing -R<grid> -F... -G<name>' on a lon/lat region silently ran the Okada deformation in Cartesian, producing wrong results with no warning. Verified against an explicit -fg run: bit-identical now. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
… 3D cube
Previously -E's Energy/Power silently overwrote the z (sea-surface) variable
in the -G 3D netCDF cube, mislabeled with z's own name/units ("Sea surface
[meters]"), while -S always wrote velocity alongside z with no way to opt
out. Add +a to both: absent, the option's quantity is the cube's only
variable (correctly named/labeled); with +a, it's written as an extra
channel next to z, matching -S's old default behavior.
Note: -S without +a now omits the z variable (previously always included).
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Esteban82
added
add-changelog
Member
|
Is there any way to check this? At least to get the same figures as the ones you’ve shown? |
Esteban82
approved these changes
Jul 8, 2026
Esteban82
left a comment
Esteban82
left a comment
Member
There was a problem hiding this comment.
Blindly approved. Let me know if you want me to test it somehow.
Member
Author
|
You can get something alike with The above created 3 grids (the Meanwhile I hit another gmt bug. 
|
Member
|
I think it works fine. I tried these commands: 
I got these messages in the terminal: |
Member
Author
|
You are plotting the bathymetry. To extract a layer from the .nc file you need #9095 That is why I used in my example |
Member
|
Ok. I see. Now I got this: 
I used these commands: |
Member
Author
|
Yes, that is correct |
Member
|
Great, now it can be merged. |
Member
|
Maybe it would be good to add some references in the rst file. For example I found this one: |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
NSWING is a non linear shallow water with nested grids tsunami modeling code. I have it for a long time compiled as a MEX file, which is used in Mirone. Now, whit help of Fable, Opus and Sonnet (credits were not enough to satisfy the hungry models) I converted it into a GMT supplement. In the transition steps Fable rewrote some parts using OpenMP and it now runs twice fast as it used to.
It's not easy to make tests for it, and the modeling result is stored in a 3D netcdf cube. GMT can be used to visualize it but a bit cumbersome. For the time being, Mirone is still the best tool I know to visualize data like this but that will, hopefully (so Claude credits help) change soon too.
The man page has a quick example of what we can get without the nesting grids and it's something like this.
I'm pasting too the result of what is known as Benchmark 1 of the Catalina set of benchmarks that compares against the analytical solution of a wave inundating a beach with 10% slope. Results are excellent.