Re: [gentoo-alt] eprefixify_patch

[Gregory M. Turner]

----- Original Message -----
> On 03-04-2012 08:37:57 -0000, Gregory M. Turner wrote:
> > Current prefix SOP is to epatch a bunch of files with
> > @GENTOO_PORTAGE_EPREFIX@ and then eprefixify the affected files.
> >  Although this adds a bit of precision and flexibility, it leads
> > to laundry lists in the ebuild, especially if/when we start
> > getting more rigorous about prefixification.  The attached eclass
> > provides eprefixify_patch, which simply prefixifies the patch
> > itself before applying.
>
> Feels dirty too ;)

Admittedly, there is a certain inelegance there.  But I can't imagine a real-world scenario which both: would correctly be solved by 'patch + eprefixify'; but would not correctly be solved by eprefixify_patch.  It would have to be something like, "well, foo.patch puts '@GENTOO_PORTAGE_EPREFIX@' into 20 different places in '${S}', but only 18 of them are to be eprefixified; the other two are correct patches to the source", which hardly seems like a use-case worth preserving.

The idea is to promote clarity and correctness by:

o Eliminating redundant information (files in patch vs. files eprefixified
  in ebuild)

o Eliminating potential for error (did we remember to eprefixify all the files
  that we patched?  did we leave behind formerly-eprefixified files in the
  ebuild?)

o Minimizing prefix-specific changes to ebuilds be getting rid of laundry-lists

o Encouraging a clear distinction between two types of patches: normal patches,
  which typically are reverse-able and have legitimate source files as both
  their input and output, and "eprefixification" patches, which leave the
  source in an intermediate state, and become pseudo-irreversible once that
  intermediate state is resolved back to a source-file (plus, this
  approach makes them reversible again).

> > It also includes some quick & dirty tools to prefixify /bin/{ba,}sh
> > shebangs.  I'm using both in my overlay.  So far, so good.  What
> > I'm using it /for/ is a more interesting question that I'd like to
> > bring up in a separate, presumably more controversial thread.
> >  Anyhow, for the moment I thought these might come in handy or at
> > least precipitate an interesting flamewar :)
>
> I fail to see how this is necessary.  Portage already fixes
> those/these shebangs, doesn't it?

Not on my system (which isn't saying much :) )... Where is that supposed to be happening?  I have a vague memory of prefix-portage doing this at one time, but the code is not jumping out at me.

-gmt