969932 - enable sendfile(2) again in libevent on freebsd

Status: RESOLVED FIXED

(Core :: IPC, defect)

Description

[Jan Beich]

Attachment: fix

Given that the typo didn't break NetBSD and OpenBSD I think it's safe to define the macro unconditionally. buffer.c already checks for platform before actually using sendfile(2).

Comment 1

[Jan Beich]

Bug 883502 had wrong comment but correct code

  ifneq (,$(filter DragonFly FreeBSD,$(OS_ARCH))) # (not OS_DRAGONFLY and not OS_FREEBSD) {

and bug 874266 converted the comment into wrong code

  if not os_freebsd and not os_dragonfly:

Comment 2

[Jan Beich]

Add a=NPOTB for aurora and beta.

Comment 3

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/integration/mozilla-inbound/rev/c0a263e2a70b

Comment 4

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/mozilla-central/rev/c0a263e2a70b

Comment 5

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/releases/mozilla-aurora/rev/c5af36178cb9
https://hg.mozilla.org/releases/mozilla-beta/rev/88806ebdaacd

Comment 6

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/releases/mozilla-b2g28_v1_3/rev/88806ebdaacd

Comment 7

[u279076]

Jan, please verify this is fixed for Firefox 28, 29 and 30. Thank you.

Comment 8

[Jan Beich]

The fix works as expected on all the branches I requested it. buffer.o (part of libevent) calls sendfile(2) defined somewhere else (in libc syscall wrapper).

$ objdump -t obj-*/ipc/chromium/buffer.o | fgrep sendfile
0000000000000000 l     O .bss   0000000000000001 use_sendfile
0000000000003ea0 g     F .text  0000000000000014 .hidden _evbuffer_testing_use_sendfile
0000000000000000         *UND*  0000000000000000 sendfile