pgrep: optimize command name copying

strncpy() implementations in glibc and musl use memset() to zero the entire destination string past the source string length. This is hot code path that runs per PID. It makes strncpy() take more cycles than anything else.

Use memccpy() instead since it never touches the destination past the string terminator.

Signed-off-by: Baruch Siach baruch.siach@siklu.com

mentioned in commit 98ca9fe3

mentioned in commit 204194cd

Nice, this looks simpler than my stzncpy proposal !118

@tt.rantala does this and !118 need work still as Jim's commits were for top only?

Let me rephrase it, I believe now this patch should be used instead of !118?

I'm currently using pgrep from procps v3.3.16 and encountering extreme slowness as described in !118 compared to my previous use of v3.2.8. Is my understanding correct that this patch provides significant improvement by not zeroing 2 MBytes of memory for each process? And that the behaviour before was much better because until v3.3.15 it only zeroed 4 KBytes for each process?

I believe this shows that the _SC_ARG_MAX value is 2 MBytes on my system.

# xargs --show-limits </dev/null 2>&1 | grep "Maximum length"
Maximum length of command we could actually use: 2094500

Is there a reason that this patch should not be applied?