[Bug dynamic-link/17645] RFE: Improve performance of dynamic loader for deeply nested DSO dependencies.

["paulo.cesar.pereira.de.andrade at gmail dot com" <sourceware-bugzilla at sourceware dot org>]

https://sourceware.org/bugzilla/show_bug.cgi?id=17645

--- Comment #9 from Paulo Andrade <paulo.cesar.pereira.de.andrade at gmail dot com> ---
Created attachment 8194
  --> https://sourceware.org/bugzilla/attachment.cgi?id=8194&action=edit
dso_deps.c - simple test implementing standalone propose sort and current glibc

Compile it as gcc -O0 -g3 dso_deps.c -o dso_deps

It accepts a very simple grammar input file, it
outputs "ORG: ..." to show how input was given,
"NEW: ..." as how the proposed patch would sort
it, and "CUR: ..." as the result of current
glibc sort. For example:

---8<---
$ cat dso_1
a b c
b c
c a
$ ./dso_deps -v dso_1
a(b c)
b(c)
c(a)
ORG: a b c
NEW: a b c
Real time: 4e-06 sec

CUR: a b c
Real time: 5e-06 sec

$ cat dso_2
a b c
d
e f
b c d e

$ ./dso_deps -v dso_2
a(b c)
b(c d e)
c()
d()
e(f)
f()
ORG: a b c d e f
NEW: a b c d e f
Real time: 5e-06 sec

CUR: a b c d e f
Real time: 3e-06 sec

$ cat dso_4
a b
b c
c d
d e
e f
$ ./dso_deps -v dso_4
a(b)
b(c)
c(d)
d(e)
e(f)
f()
ORG: a b c d e f
NEW: a b c d e f
Real time: 5e-06 sec

CUR: a b c d e f
Real time: 3e-06 sec

$ cat dso_5
b a
c b
d c
e d
f e
$ ./dso_deps -v dso_5
b(a)
a()
c(b)
d(c)
e(d)
f(e)
ORG: b a c d e f
NEW: f e d c b a
Real time: 5e-06 sec

CUR: f e d c b a
Real time: 6e-06 sec

---8<---


This is an example matching the current glibc build
time test, and that the proposed patch has a different
result from current glibc:
---8<---
$ cat dso_6
a2 a1
b2 b1 a2
a3 b2 b1
a4 a3
main a4 a1 b2
$ ./dso_deps -v dso_6
a2(a1)
a1()
b2(b1 a2)
b1()
a3(b2 b1)
a4(a3)
main(a4 a1 b2)
ORG: a2 a1 b2 b1 a3 a4 main
NEW: main a4 a3 b2 a2 a1 b1
Real time: 6e-06 sec

CUR: main a4 a3 b2 b1 a2 a1
Real time: 7e-06 sec

---8<---

-- 
You are receiving this mail because:
You are on the CC list for the bug.