Re: false alert on new pkg_add shlibs check?

[Thomas Klausner <wiz%gatalith.at@localhost>]

On Fri, Oct 18, 2024 at 06:32:32PM -0400, Greg Troxel wrote:
> Thomas Klausner <wiz%gatalith.at@localhost> writes:
> 
> > On Thu, Oct 17, 2024 at 11:23:02AM -0400, Greg Troxel wrote:
> >> Sorry, I meant to change the second one to "it's a bug that -f doesn't
> >> bypass this check".  I don't see how anybody can get past this, and I
> >> don't think we can argue there won't be bugs in checks, or that people
> >> won't need to use buggy packages.
> >
> > Since this additional check was implemented because of a bug report
> > that caused a pkg_add core dump, I'm against allowing -f to override
> > this.
> >
> > The problem is that the package might have an INSTALL that runs a
> > program that actually needs the REQUIRED library.
> 
> But the core dump is due to (a bug really but) INSTALL ending
> abnormally, and that  can happen from many causes, and this REQUIRED bit
> is about some other set of causes.  It's plausible that
>   1) INSTALL will go badly even if all libs are ok
>   2) INSTALL can be fine even if there is a lib problem
>   3) the libs are actually ok but the metadata is wrong or the check is buggy
> 
> > Probably one(TM) could also improve that part so that pkg_add doesn't
> > dump core when an INSTALL script fails to run completely, but that's
> > not implemented yet and before that I'm against adding the override.
> 
> Becaues a user having the ability to override, which might provoke a
> core dump, is super terrible?

Yes.

> And that not being able to install a
> package, even if a little messed up but working, is better?

Yes.

> With the new check, there's an error, and if someone adds -f and gets
> pkg_add.core, that's not really hard to figure out.   Vs having to edit
> the sources or use some older version.

Or editing the broken binary package's meta data.
Or fixing the binary package.

I think fixing the binary package at the pkgsrc level is the right
solution.

Making pkg_install behave better for other INSTALL script problems is
a worthy goal too, of course.
 Thomas

(not with my pmc hat on)