[oi-dev] shell/ksh93

[Jim Klimov]

On March 30, 2021 5:57:11 PM UTC, Gary Mills <gary_mills at fastmail.fm> wrote:
>On Tue, Mar 30, 2021 at 05:53:00PM +0200, stes at PANDORA.BE wrote:
>> 
>> I think there is no need to do bulk conversion of OpenIndiana
>> packages, due to the change (the commit
>> c063eb990f530561e469b3c1e4bb64230456c0da in illumos-gate).
>
>> Older versions of packages that depend on SUNWcs (core solaris or
>> core system) continue to work fine.
>
>> New versions of packages can modify their manifest and Makefile and
>> require pkg:/shell/ksh93.
>
>This analysis seems correct, at least for installed systems.  We do
>have to think about updates to installed systems, new installations,
>and the build system, however.
>
>As long as the ksh package is a dependancy of another package that is
>always installed, it will always be installed as well.  That situation
>should be fine for updates and new installations.
>
>Certainly the OI build system will be affected.  I'm assuming that
>this system only rebuilds packages when something in them has changed.
>So, some of those packages will fail to build, because they do not
>have the new ksh package in the REQUIRED_PACKAGES gmake variable.  The
>easiest course might be to fix the Makefile of various packages as the
>rebuild breaks.  An alternative might be to identify the packages and
>fix them in advance, so that the rebuild succeeds.  I don't think that
>a permanent global fix is possible for the build system.

Hi all,

I currently presume (did not check to prove) that SUNWcs might (for some transition period or forever) depend on shell/ksh93 and so package builds with unchanged recipes would succeed. Just two packages instead of one, delivering more or less same bytes.

Updating recipes may be more due to optimization (do we really need the bulk of OS/Net in the scratch build zone to build this script-only package), if SUNWcs is/will-be-one-day at all avoidable as an installation footprint, and/or if an older version may be deployed and unchanged since golden image because it is not a relevant critical part of the recipe requirements? 

And possibly limit the impact of inverse dependencies (we critically changed a low-level dependency so everyone using them should get rebuilt - the attack surface of ksh93 alone is smaller than SUNWcs altogether).

No idea if there's a grain of practical truth to my ideas above, just that's what I think about it today, based on experience with various OSes and build systems, until proven wrong :)

Jim



--
Typos courtesy of K-9 Mail on my Android