Re: Documentation of the TIMEFMT variable

[Jun T <takimoto-j@xxxxxxxxxxxxxxxxx>]

> 2023/03/24 23:43、Oliver Kiddle <opk@xxxxxxx>のメール:

>> Is there anyone who can confirm that ru_maxrss is set to zero?
>> If it is always zero, then we need not bother with page to KB conversion.
> 
> It does appear to always be zero.

Thanks. I hope the following is enough:


diff --git a/Src/jobs.c b/Src/jobs.c
index 59ddd952e..4d7172550 100644
--- a/Src/jobs.c
+++ b/Src/jobs.c
@@ -888,8 +888,13 @@ printtime(struct timeval *real, child_times_t *ti, char *desc)
 		break;
 #endif
 #ifdef HAVE_STRUCT_RUSAGE_RU_MAXRSS
+#ifdef RU_MAXRSS_IS_IN_BYTES
+# define MAXRSS_IN_KB(x) ((x)/1024)
+#else
+# define MAXRSS_IN_KB(x) (x)
+#endif
 	    case 'M':
-		fprintf(stderr, "%ld", ti->ru_maxrss / 1024);
+		fprintf(stderr, "%ld", MAXRSS_IN_KB(ti->ru_maxrss));
 		break;
 #endif
 #ifdef HAVE_STRUCT_RUSAGE_RU_MAJFLT
@@ -1036,7 +1041,7 @@ should_report_time(Job j)
 
 #ifdef HAVE_GETRUSAGE
     if (reportmemory >= 0 &&
-	j->procs->ti.ru_maxrss / 1024 > reportmemory)
+	MAXRSS_IN_KB(j->procs->ti.ru_maxrss) > reportmemory)
 	return 1;
 #endif
 
diff --git a/configure.ac b/configure.ac
index f340d2993..e6ced85d9 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1965,6 +1965,15 @@ if test x$ac_cv_func_getrusage = xyes; then
 #endif
 #include <sys/resource.h>])
 fi
+dnl On some OSes (only macOS?) ru_maxrss is in bytes (not in kilobytes).
+dnl Solaris uses pages as the unit, but ru_maxrss is set to zero anyway.
+AH_TEMPLATE(RU_MAXRSS_IS_IN_BYTES,
+[Define to 1 if ru_maxrss in struct rusage is in bytes.])
+case "$host_os" in
+  darwin*)
+    AC_DEFINE(RU_MAXRSS_IS_IN_BYTES)
+  ;;
+esac
 
 
 dnl --------------------------------------------