fix(mem): reconcile peak RSS with live current RSS (Linux statm vs ru_maxrss)

cbm_mem_peak_rss() returned mimalloc's peak_rss (from getrusage ru_maxrss,
KB-granular) while on Linux cbm_mem_rss() reads the live /proc/self/statm
value (page-granular, the RSS-undercount fallback). The two sources can
disagree by a few pages, so a live current read could momentarily exceed the
reported peak and break the definitional peak >= current invariant — flaking
the mem_peak_rss_gte_rss test on the Linux/ARM CI leg (observed as peak 172KB
under current). Reconcile the sources so the reported peak is at least the
current RSS. Not observable on macOS, where both come from mimalloc.

The mem_peak_rss_gte_rss guard now faults in a 32MB buffer before asserting so
the invariant is checked against a non-trivial live current read.

Signed-off-by: Martin Vogel <martin.vogel.tech@gmail.com>

Author: DeusData

src/foundation/mem.c

@@ -212,6 +212,17 @@ size_t cbm_mem_rss(void) {
 size_t cbm_mem_peak_rss(void) {
     size_t peak_rss = 0;
     mi_process_info(NULL, NULL, NULL, NULL, &peak_rss, NULL, NULL, NULL);
+    /* Peak RSS is by definition >= current RSS. On Linux cbm_mem_rss() reads the
+     * live /proc/self/statm value (page-granular), while mimalloc's peak_rss
+     * comes from getrusage's ru_maxrss (KB-granular, and it can lag the live
+     * statm read by a few pages). Reading the two sources independently lets a
+     * fresh current read momentarily exceed the reported peak, breaking the
+     * peak >= current invariant. Reconcile them: the true peak is at least the
+     * current RSS. (Not observable on macOS, where both come from mimalloc.) */
+    size_t current = cbm_mem_rss();
+    if (current > peak_rss) {
+        peak_rss = current;
+    }
     if (peak_rss > 0) {
         return peak_rss;
     }

tests/test_mem.c

@@ -180,10 +180,23 @@ TEST(mem_rss_positive) {
 
 TEST(mem_peak_rss_gte_rss) {
     cbm_mem_init(0.5);
+    /* peak >= current RSS is definitional. Regression guard for the Linux
+     * statm-vs-ru_maxrss source mismatch: cbm_mem_rss() reads the live
+     * /proc/self/statm value (page-granular) while mimalloc's peak comes from
+     * getrusage ru_maxrss (KB-granular, and it lags), so a live current read
+     * could momentarily exceed the reported peak by a few pages and break the
+     * invariant. cbm_mem_peak_rss() now reconciles the two sources. Touch a
+     * fresh buffer so the check runs against a non-trivial live current read.
+     * (Linux-only bug — macOS reads both from mimalloc; it flaked on the
+     * Linux/ARM CI leg, which is the authoritative reproduction tier.) */
+    size_t n = 32 * 1024 * 1024;
+    char *p = (char *)malloc(n);
+    ASSERT_NOT_NULL(p);
+    memset(p, 0xBE, n); /* fault in all pages so current RSS is non-trivial */
     size_t rss = cbm_mem_rss();
     size_t peak = cbm_mem_peak_rss();
-    /* Peak must be >= current RSS */
     ASSERT_GTE(peak, rss);
+    free(p);
     PASS();
 }