Date: Fri, 21 Sep 2012 07:58:18 -0500 From: Bryan Drewery <bdrewery@freebsd.org> To: Beat Gaetzi <beat@freebsd.org>, ports-committers@freebsd.org, svn-ports-all@freebsd.org, svn-ports-head@freebsd.org, poyopoyo@puripuri.plala.or.jp Subject: Re: svn commit: r304594 - head/Mk Message-ID: <505C646A.6030009@FreeBSD.org> In-Reply-To: <868vc3rhqm.wl%poyopoyo@puripuri.plala.or.jp> References: <201209201951.q8KJpUNr085151@svn.freebsd.org> <868vc3rhqm.wl%poyopoyo@puripuri.plala.or.jp>
next in thread | previous in thread | raw e-mail | index | archive | help
On 9/21/2012 1:51 AM, poyopoyo@puripuri.plala.or.jp wrote:
> At Thu, 20 Sep 2012 19:51:30 +0000 (UTC),
> Beat Gaetzi wrote:
>> - Make 'do-package' atomic while creating the pkg file [4]
>
>> Modified: head/Mk/bsd.port.mk
>> ==============================================================================
>> @@ -3911,11 +3905,15 @@ do-package: ${TMPPLIST}
>> fi; \
>> fi; \
>> fi
>> - @if ${PKG_CMD} -b ${PKGNAME} ${PKGFILE}; then \
>> + @TMPPKGFILE=$$(mktemp -t pkg); \
>> + trap "${RM} -f $${TMPPKGFILE} $${TMPPKGFILE}${PKG_SUFX}; exit 1" 1 2 3 5 10 13 15; \
>> + if ${PKG_CMD} -b ${PKGNAME} $${TMPPKGFILE}; then \
>> + ${MV} -f $${TMPPKGFILE}${PKG_SUFX} ${PKGFILE}; \
>> if [ -d ${PACKAGES} ]; then \
>> cd ${.CURDIR} && eval ${MAKE} package-links; \
>> fi; \
>> else \
>> + ${RM} -f $${TMPPKGFILE}; \
>> cd ${.CURDIR} && eval ${MAKE} delete-package; \
>> exit 1; \
>> fi
>
> This does not seem to atomic at all. Well being visible when PKG_SUFX is set.
How is a single mv(1) on a tempfile not atomic? The purpose here is to
avoid the very long pkg_create(1) call and avoid leaving an invalid
package behind on ctrl+c or other interruptions.
>
> ==
> # make package -C /usr/ports/ports-mgmt/portlint PKG_SUFX=.tgz
> ...
> | ===> Registering installation for portlint-2.13.12
> | ===> Building package for portlint-2.13.12
> | mv: rename /home/tmp/pkg.ENocPBwh.tgz to /home/packages/10/All/portlint-2.13.12.tgz: No such file or directory
> | *** [do-package] Error code 1
> |
> | Stop in /usr/ports/ports-mgmt/portlint.
>
> # ls -l /home/tmp/pkg.ENocPBwh*
> -rw------- 1 root wheel 0 Sep 21 14:36 /home/tmp/pkg.ENocPBwh
> -rw-r--r-- 1 root wheel 32270 Sep 21 14:36 /home/tmp/pkg.ENocPBwh.tbz
> ==
>
> (1) pkg_create creates package with default suffix *added* to
> given pkg-filename if pkg-filename does not have one of known suffixes
> which are now .tgz, .tbz and .txz.
> (2) TMPPKGFILE does not give PKG_SUFX to mktemp. This makes pkg_create
> *adding* known default suffix to given TMPPKGFILE, this time
> "pkg.ENocPBwh.tbz".
> (3-a) when PKG_SUFX is set and it was other than default suffix pkg_create
> knows, make package fails.
> (3-b) when PKG_SUFX is not set, it looks like make package successfully
> completes, but its operation is not atomic at all, because intended
> TMPPKGFILE did not used at all (see empty file above, "pkg.ENocPBwh".)
> (4) TMPPKGFILE is left on TMPDIR and not removed.
>
> An hour of effort make me giving up fixing this correctly -- are someone
> able to tell mktemp to use pkg.XXXXXXXX.tgz as template? I could not. :-)
>
> Anyway, this chunk of patch does not work as intended currently and
> breaks PKG_SUFX support badly. It would be nice to be backed out until
> someone makes this really work, PKG_SUFX compliant and being atomic.
I will look at this.
>
--
Regards,
Bryan Drewery
bdrewery@freenode/EFNet
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?505C646A.6030009>