svn commit: r394871 - in head: Mk mail/thunderbird mail/thunderbird/files www/firefox www/firefox-esr www/firefox-esr/files www/firefox/files www/libxul www/libxul/files www/seamonkey www/seamonkey...

John Marino freebsd.contact at marino.st
Thu Aug 20 15:26:16 UTC 2015


On 8/20/2015 5:15 PM, Jan Beich wrote:
> John Marino <freebsd.contact at marino.st> writes:
>> By switching the comparison from < to >=, I think this just broke
>> DragonFly which does not support jemalloc on any version.
> 
> Ahem, OSVERSION=9999999 in DPorts' bsd.port.mk isn't sane either.
> Either emulate it (along with OSREL) via USES=alias, derive from
> __DragonFly_version or force breakage by not defining it.

I disagree.  Having it work 90% of the time by default is better than 0%
of the time by default.  OSVERSION only works for FreeBSD, period.
there is no emulation possible.  We have DFLYVERSION analog but that's
not known in the stock version.

With correct usage, OSVERSION is never used without OPSYS.  Those are
coupled data points.  I "get" the argument that some people have that
OPSYS should be assumed to be "FreeBSD" and thus doesn't need to be
checked, but obviously that is not conducive to DragonFly support.


> 
> --enable-jemalloc is a toggle for bundled jemalloc. Without it the build
> defaults to system malloc. configure also checks if system malloc
> supports jemalloc non-standard API. The API is mainly used for
> about:memory. However, lack of malloc_usable_size() may impact
> performance but it is usually available elsewhere (musl, glibc, etc).
> 
> files/extra-patch-bug1125514 should be a nop for DragonFly, FreeBSD 9.x
> and FreeBSD 10.x. It can cause garbage in memory stats for bundled jemalloc
> but otherwise explicitly separated by make(1) conditional.
> 
> My intent is to make DragonFly use bundled jemalloc like older releases
> of FreeBSD. Try applying below commit under ${WRKSRC}/memory/jemalloc.


Why?  jemalloc has never worked on DF.  Ever port with jemalloc support,
bundled or otherwise fails.



> To avoid undefined behavior OPSYS check has to be added twice for (large
> and small) OSVERSION. Or maybe inverse it like

or just the entire section in "If FreeBSD".
If you don't want to adjust then we'll just have to carry another patch
in the overlay.  I'm not going to push the issue.

Thanks,
John



More information about the svn-ports-head mailing list