Did buildinstall.functions (and/or mkinitrd), which is now written
in bash, used to be written in some other language, such as python?
get_dso_deps contains a comment "I still hate shell."
I have been building and testing a DVD from rawhide daily using
pungi. Building takes about 40 to 45 minutes: 10 minutes user CPU,
14 minutes system CPU, and 16 to 20 minutes download.
I'd like that to be faster, say 8 to 10 minutes plus download.
After Bug 559647 (quadratic work: genisoimage per .rpm) then
the next longest piece is upd-instroot. "strace -f -c" shows
that there are over 125,000 successful execve(). Now install.img
has around 14,600 files, so the average is more than 8 execve
per file. A detailed examination using strace shows over 97,000
execve by upd-instroot:
34573 upd-instroot execve("/usr/bin/dirname",
15167 upd-instroot execve("/bin/cp",
12513 upd-instroot execve("/bin/egrep",
11478 upd-instroot execve("/usr/bin/file",
5538 upd-instroot execve("/bin/sed",
5010 upd-instroot execve("/bin/basename",
3675 upd-instroot execve("/usr/bin/readlink",
3090 upd-instroot execve("/usr/sbin/chroot",
2264 upd-instroot execve("/bin/rm",
1134 upd-instroot execve("/bin/ln",
1046 upd-instroot execve("/bin/grep",
1031 upd-instroot execve("/lib64/ld-2.11.90.so",
1030 upd-instroot execve("/bin/uname",
703 upd-instroot execve("/usr/bin/find",
627 upd-instroot execve("/bin/mkdir",
[snip]
and the culprits are instFile() and get_dso_deps() and callers.
One process making some direct calls per file, instead of
clone+execve several times per file, should be much faster.
--
_______________________________________________
Anaconda-devel-list mailing list
Anaconda-devel-list@xxxxxxxxxx
https://www.redhat.com/mailman/listinfo/anaconda-devel-list
