On Sun, 23 May 2010 17:19:03 +0100 Owain Ainsworth <[email protected]> wrote: > On Sun, May 23, 2010 at 10:15:21AM -0600, Ted Roby wrote: > > On Sun, May 23, 2010 at 9:55 AM, Owain Ainsworth > > <[email protected]>wrote: > > > > > > > > > > > > > > You mean that the build failed? or that the corruption happened? > > > > > > > > I mean the Xenocara and kernel builds fail, depending. > > I have used 'log' to replace i915_drv.h when it was changed as well. > > to my knowledge the kernel should always have build during that range. > This means that you have done something wrong. > > cvs up -D "<date before/after commit>" in sys/dev/pci/drm should be > sufficient. >
Unfortunately, by default `cvs up` will try to merge local changes with whatever it gets from the repository. Since the '-C' flag is broken on all mirrors running opencvs (see bug user/6363), trying to use '-C' to abandon/overwrite local changes will not always work. You have to find a mirror running gnucvs like [email protected]:/cvs for the '-C' flag to work as expected. Additionally, the '-D <date_spec>' flag is sticky, so if you ever used it before, it will make a real mess unless cleared with the '-A' flag. You'd want something like this: # cd /usr # cvs -d$CVSROOT up -ACPd -D "2010-04-24 23:59 UTC" xenocara But, even the above may not do what you want (see below) > > I got lucky catching the April 24 snapshot, and > > having the new May 22 snapshot posted. > > > > With all the lib changes I have had troubles doing unsupported > > things like downgrades, or make builds from my own cvs copy. > > > > > With two variables in play that doesn't tell me anywhere near as > > > much as I would need... > > > > > > > Please advise if I'm missing some other way to get a > > Xenocara build from May 9, 10 and 11. > > cd /usr/xenocara > cvs up -D "2010/05/9 <your time here>" > make bootstrap > make obj > make build > There is something wonky with the CVS, tested with mirrors running both opencvs and the one I know is running gnucvs. I'm still trying to figure out the cause.... Given a fresh install of Apr 15 snapshot, followed by updating kernel and userland to Apr 25 via src, the following fails: # rm -rf /usr/xobj/* # rm -fr /usr/xenocara/* # cd /usr # cvs -d$CVSROOT co -D "2010-04-18 23:59 UTC" xenocara # cd /usr/xenocara # make bootstrap && make obj && make build It should work but it doesn't. The build dies due to incorrect headers and/or defines in: /usr/xenocara/lib/libX11/modules/im/ximcp/imLcFlt.c or /usr/xenocara/lib/libX11/modules/im/ximcp/imLcLkup.c If I manually fix the missing `#define XK_BRAILLE` the build just dies elsewhere. As for the potential of having bad stuff installed in /usr/X11R6/, I checked it, but I also untarred the snapshot x*.tgz files to be sure and a second attempted build (with clean xobj) also failed. The problem is the source that's getting checked out is wrong. According to the description of the date_spec in cvs(1), I've got the ISO date format right, but in trying to debug this I've even tried "GMT" rather than "UTC" and it still fails. Either with either the above cvs checkout command, or with the above cvs update command (-ACPd -D <date_spec>), I've tested the following dates of xenocara and they all fail to build. 2010-04-17 23:59 UTC 2010-04-18 23:59 UTC 2010-04-19 23:59 UTC 2010-04-20 23:59 UTC 2010-04-21 23:59 UTC 2010-04-22 23:59 GMT 2010-04-23 23:59 UTC 2010-04-24 23:59 UTC 2010-04-25 23:59 UTC I've tried three different mirrors. [email protected]:/cvs (gnucvs) [email protected]:/cvs (opencvs) [email protected]:/cvs (opencvs) On the other hand, if I extract my personal backup archive of xenocara as updated on Apr 18th, it builds perfectly. My personal backup archive of xenocara updated on the 25th also builds correctly. The only way I'll be able to pick up my "Not Me" flag and wave it with pure confidence is if I did another fresh install of the Apr 15 snap, and repeat the cvs checkout of Apr 18 xenocara as above. If that doesn't build, then I can be certain the '-D' flag is busted, but at the moment, it seems the '-D' flag being broken is highly likely. Since it seems I'm the first to find a problem with the '-D' flag, I'm testing to see if it is being caused by my using the ISO date_spec format by doing a similar checkout with the "Internet" date_spec format. i.e. # rm -rf /usr/xobj/* # rm -fr /usr/xenocara/* # cd /usr # cvs -d$CVSROOT co -D'18 Apr 2010 10:30 GMT' xenocara If the above works, then there's a problem in parsing the ISO date_spec as stated in the cvs(8) man page. jcr
