Filter simulation CSV to comparison signals only

src/compare.jl

@@ -334,7 +334,7 @@ end
 
 """
     compare_with_reference(sol, ref_csv_path, model_dir, model;
-                           settings) → (total, pass, skip, diff_csv)
+                           settings, signals) → (total, pass, skip, diff_csv)
 
 Compare a DifferentialEquations / MTK solution against the MAP-LIB reference CSV.
 
@@ -354,33 +354,42 @@ is written whenever there are failures or skipped signals.
 
 # Keyword arguments
 - `settings` — a `CompareSettings` instance controlling tolerances and the
-               error function.  Defaults to the module-level settings returned
-               by `compare_settings()`.  Use `configure_comparison!` to change
-               the defaults, or pass a local `CompareSettings(...)` here.
+               error function.
+- `signals`  — explicit list of signal names to compare.  When non-empty,
+               overrides `comparisonSignals.txt` and the full reference CSV
+               column list.
 """
 function compare_with_reference(
     sol,
     ref_csv_path::String,
     model_dir::String,
     model::String;
     settings::CompareSettings = _CMP_SETTINGS,
+    signals::Vector{String}   = String[],
 )::Tuple{Int,Int,Int,String}
 
     times, ref_data = _read_ref_csv(ref_csv_path)
     isempty(times) && return 0, 0, 0, ""
 
-    # Determine which signals to compare: prefer comparisonSignals.txt
-    sig_file           = joinpath(dirname(ref_csv_path), "comparisonSignals.txt")
-    using_sig_file     = isfile(sig_file)
-    signals = if using_sig_file
-        sigs = filter(s -> lowercase(s) != "time" && !isempty(s), strip.(readlines(sig_file)))
-        sigs_missing = filter(s -> !haskey(ref_data, s), sigs)
-        isempty(sigs_missing) || error("Signal(s) listed in comparisonSignals.txt not present in reference CSV: $(join(sigs_missing, ", "))")
-        sigs
+    # Determine which signals to compare.
+    # Prefer the caller-supplied list; fall back to comparisonSignals.txt, then
+    # all columns in the reference CSV.
+    signals = if !isempty(signals)
+        sigs_missing = filter(s -> !haskey(ref_data, s), signals)
+        isempty(sigs_missing) || error("Signal(s) not present in reference CSV: $(join(sigs_missing, ", "))")
+        signals
     else
-        filter(k -> lowercase(k) != "time", collect(keys(ref_data)))
+        sig_file = joinpath(dirname(ref_csv_path), "comparisonSignals.txt")
+        if isfile(sig_file)
+            sigs = filter(s -> lowercase(s) != "time" && !isempty(s), strip.(readlines(sig_file)))
+            sigs_missing = filter(s -> !haskey(ref_data, s), sigs)
+            isempty(sigs_missing) || error("Signal(s) listed in comparisonSignals.txt not present in reference CSV: $(join(sigs_missing, ", "))")
+            sigs
+        else
+            filter(k -> lowercase(k) != "time", collect(keys(ref_data)))
+        end
     end
-    n_total     = length(signals)
+    n_total = length(signals)
 
     # ── Build variable accessor map ──────────────────────────────────────────────
     # var_access: normalized name → Int (state index) or MTK symbolic (observed).

src/pipeline.jl

@@ -76,20 +76,34 @@ function test_model(omc::OMJulia.OMCSession, model::String, results_root::String
     par_ok || return ModelResult(
         model, true, exp_t, exp_err, false, par_t, par_err, false, 0.0, "", 0, 0, 0, "")
 
+    # Resolve reference CSV and comparison signals early so phase 3 can filter
+    # the CSV output to only the signals that will actually be verified.
+    ref_csv     = isempty(ref_root) ? nothing : _ref_csv_path(ref_root, model)
+    cmp_signals = if ref_csv !== nothing
+        sig_file = joinpath(dirname(ref_csv), "comparisonSignals.txt")
+        if isfile(sig_file)
+            String.(filter(s -> lowercase(s) != "time" && !isempty(s), strip.(readlines(sig_file))))
+        else
+            _, ref_data = _read_ref_csv(ref_csv)
+            filter(k -> lowercase(k) != "time", collect(keys(ref_data)))
+        end
+    else
+        String[]
+    end
+
     # Phase 3 ──────────────────────────────────────────────────────────────────
-    sim_ok, sim_t, sim_err, sol = run_simulate(ode_prob, model_dir, model; csv_max_size_mb)
+    sim_ok, sim_t, sim_err, sol = run_simulate(ode_prob, model_dir, model;
+                                               csv_max_size_mb, cmp_signals)
 
     # Phase 4 (optional) ───────────────────────────────────────────────────────
     cmp_total, cmp_pass, cmp_skip, cmp_csv = 0, 0, 0, ""
-    if sim_ok && !isempty(ref_root)
-        ref_csv = _ref_csv_path(ref_root, model)
-        if ref_csv !== nothing
-            try
-                cmp_total, cmp_pass, cmp_skip, cmp_csv =
-                    compare_with_reference(sol, ref_csv, model_dir, model)
-            catch e
-                @warn "Reference comparison failed for $model: $(sprint(showerror, e))"
-            end
+    if sim_ok && ref_csv !== nothing
+        try
+            cmp_total, cmp_pass, cmp_skip, cmp_csv =
+                compare_with_reference(sol, ref_csv, model_dir, model;
+                                       signals = cmp_signals)
+        catch e
+            @warn "Reference comparison failed for $model: $(sprint(showerror, e))"
         end
     end
 

src/simulate.jl

@@ -6,18 +6,24 @@ import ModelingToolkit
 import Printf: @sprintf
 
 """
-    run_simulate(ode_prob, model_dir, model; csv_max_size_mb) → (success, time, error, sol)
+    run_simulate(ode_prob, model_dir, model; cmp_signals, csv_max_size_mb) → (success, time, error, sol)
 
 Solve `ode_prob` with Rodas5P (stiff solver).  On success, also writes the
-full solution as a CSV file `<Short>_sim.csv` in `model_dir`.
+solution as a CSV file `<Short>_sim.csv` in `model_dir`.
 Writes a `<model>_sim.log` file in `model_dir`.
 Returns `nothing` as the fourth element on failure.
 
-CSV files larger than `csv_max_size_mb` MiB are deleted and replaced with a
+When `cmp_signals` is non-empty, only observed variables whose names appear in
+that list are written to the CSV, keeping file sizes small when only a subset
+of signals will be compared.
+
+CSV files larger than `csv_max_size_mb` MiB are replaced with a
 `<Short>_sim.csv.toobig` marker so that the report can note the omission.
 """
-function run_simulate(ode_prob, model_dir::String, model::String;
-                      csv_max_size_mb::Int = CSV_MAX_SIZE_MB)::Tuple{Bool,Float64,String,Any}
+function run_simulate(ode_prob, model_dir::String,
+                      model::String;
+                      cmp_signals    ::Vector{String} = String[],
+                      csv_max_size_mb::Int            = CSV_MAX_SIZE_MB)::Tuple{Bool,Float64,String,Any}
     sim_success = false
     sim_time    = 0.0
     sim_error   = ""
@@ -67,7 +73,13 @@ function run_simulate(ode_prob, model_dir::String, model::String;
             sys      = sol.prob.f.sys
             vars     = ModelingToolkit.unknowns(sys)
             obs_eqs  = ModelingToolkit.observed(sys)
-            obs_syms = [eq.lhs for eq in obs_eqs]
+            # Only save observed variables that appear in cmp_signals.
+            # This avoids writing thousands of algebraic variables to disk when
+            # only a handful are actually verified during comparison.
+            norm_cmp = Set(_normalize_var(s) for s in cmp_signals)
+            obs_eqs_filtered = isempty(norm_cmp) ? obs_eqs :
+                filter(eq -> _normalize_var(string(eq.lhs)) in norm_cmp, obs_eqs)
+            obs_syms = [eq.lhs for eq in obs_eqs_filtered]
             col_names = vcat(
                 [_clean_var_name(string(v)) for v in vars],
                 [_clean_var_name(string(s)) for s in obs_syms],