This is the mail archive of the cygwin-apps mailing list for the Cygwin project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [PATCH] setup: fix abnormal exit test for postinstall scripts


On Thu, 16 Mar 2006, Max Bowsher wrote:

> Igor Peshansky wrote:
> > On Tue, 14 Mar 2006, Max Bowsher wrote:
> >
> >> Igor Peshansky wrote:
> >>> On Thu, 9 Mar 2006, Dave Korn wrote:
> >>>
> >>>> On 09 March 2006 23:14, Max Bowsher wrote:
> >>>>
> >>>>>>> 	* script.cc (Script::run): Fix inverted test for abnormal exit.
> >>>>>>> 	Do not rename to ".done" unless completed successfully.
> >>>>>> And ping (attached as "setup-script-exit-code-fix.patch").
> >>>>> Do we necessarily want to try to re-run failed scripts the next
> >>>>> time setup installs some packages?
> >>> Why not?  It would make recovery from a hosed install because of
> >>> in-use DLLs easy enough -- just re-run setup and select "Keep",
> >>> which will only rerun the postinstall scripts.

Speaking of in-use DLLs, did you have a chance to look at
<http://cygwin.com/ml/cygwin-apps/2006-03/msg00060.html>?

> >>>>> Perhaps renaming to ".failed" would be better that not renaming.
> >>>>>
> >>>>> Max.
> >>>>   Perhaps setup should check when you first start it up whether
> >>>> there are any postinstall scripts left lying around from last time
> >>>> and offer to run them for you then and there?  Failed postinstalls
> >>>> should be run to completion *before* next updating the package!
> >>>
> >>> Why?  I'm not so certain.  So your preremove will fail -- who cares,
> >>> it would also fail if "cygwin" is upgraded and is uninstalled before
> >>> the preremove script is run.  Next time you upgrade, it'll be like
> >>> the initial install all over again, and the *new* postinstall will
> >>> run.  If you didn't touch the package, however, the postinstall that
> >>> failed before will be re-run.  If it failed because of something in
> >>> the environment when setup was run (e.g., a stale DLL in memory), it
> >>> will now succeed and will be renamed to .done.  If it fails again,
> >>> we'll know something was wrong, and will recommend looking at the
> >>> output in setup.log.full.
> >>
> >> I'm concerned about introducing weird subtle edge cases. For example:
> >>
> >> Upgrade package in which old version has preremove, new version does
> >> not. preremove fails. Next time setup runs, stale preremove zaps parts
> >> of the upgraded package.
> >>
> >> Granted, it's a fairly tenuous situation, but at least the current
> >> behaviour is very predictable: scripts are only re-run manually.
> >
> > Umm, we can never have stale postinstalls and preremoves, since those
> > are part of the manifest (pkg.lst) and will be removed by the
> > "uninstall" part of the upgrade.  The only way we can have a stale
> > postinstall is if it no longer has the name in the manifest (i.e., is
> > renamed to ".done").  Am I missing something?
>
> Ah, I hadn't thought of that.
>
> So, I think we should always rename preremove scripts, because we
> certainly don't want a failed preremove script to be removed by the
> later file-removal phase - it might be wanted for debugging.

I would even go one step further, and cancel the uninstall of the current
package and all packages it depends on if the preremove failed.  But
that should be a separate patch.

> As for postinstall scripts ... I think ideally it would be a separate
> operation (c.f. 'dpkg --configure --pending'), but I guess we can go
> with the simple solution for now, and defer a more complex solution
> until someone has the inclination.

So, does this mean "please check in"?
	Igor
-- 
				http://cs.nyu.edu/~pechtcha/
      |\      _,,,---,,_	    pechtcha@cs.nyu.edu | igor@watson.ibm.com
ZZZzz /,`.-'`'    -.  ;-;;,_		Igor Peshansky, Ph.D. (name changed!)
     |,4-  ) )-,_. ,\ (  `'-'		old name: Igor Pechtchanski
    '---''(_/--'  `-'\_) fL	a.k.a JaguaR-R-R-r-r-r-.-.-.  Meow!

"Las! je suis sot... -Mais non, tu ne l'es pas, puisque tu t'en rends compte."
"But no -- you are no fool; you call yourself a fool, there's proof enough in
that!" -- Rostand, "Cyrano de Bergerac"


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]