Hello Warner,
Den 09/11/2012 kl. 15.36 skrev Warner Losh <[email protected]>:
> On Nov 9, 2012, at 3:52 AM, Erik Cederstrand wrote:
>
>> Hello toolchainers,
>>
>> I'm attempting to clean up hardcoded ar(1) flags in the tree to use the
>> global ARFLAGS in share/mk/sys.mk instead. I want to be able to add "-D" to
>> ARFLAGS and have it used everywhere.
>>
>> The patch changes some hardcoded flags from e.g. "cru" to the default "rl"
>> or "rv" from sys.mk. Looking at the manpage for ar(1), I'm pretty sure this
>> is safe, and my runtime tests haven't turned out any problems. Loosing the
>> "u" flags means loosing a bit in performance in theory, but I have tested
>> this to be negligible in a buildworld / kernel run. In a later iteration,
>> maybe the default flags can be added a "u".
>>
>> Are there any problems with this patch?
>
> I don't like losing the 'c' flag. Makes things in the build too whiny.
> Why purposely lose the 'u' flag that you know helps performance?
> Why move from cq to rl? This can be a big slow down...
I'd actually like to add both 'u' and 'c', I just didn't want to both clean up
and change the default in the same patch. A followup patch could be:
Index: share/mk/sys.mk
===================================================================
--- share/mk/sys.mk (revision 242822)
+++ share/mk/sys.mk (working copy)
@@ -39,7 +39,7 @@
.if defined(%POSIX)
ARFLAGS ?= -rv
.else
-ARFLAGS ?= rl
+ARFLAGS ?= cru
.endif
RANLIB ?= ranlib
('l' is obsolete).
> What data can you offer that the buildworld run time is negligible? Even on
> slower platforms?
I don't have any slower platforms available, at least not your definition of
slower :-)
Erik
_______________________________________________
[email protected] mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-toolchain
To unsubscribe, send any mail to "[email protected]"
