Re: [gentoo-alt] Prefix bootstrap stage3 pinentry fails when host system has FLTK

[Stephen McCamant]

>>>>> "FG" == Fabian Groffen <grobian@g.o> writes:

FG> Hi Stephen,
FG> pinentry indeed has an "automagic dependency", the USE-flag would
FG> solve this (be it that the patch you mention actually misses the
FG> dependency when USE=fltk is in effect).

FG> There is no easy way for you to fix it, other than faking it:
FG> create a dummy $EPREFIX/usr/bin/fltk-config that returns false,
FG> e.g. using
FG> #!/usr/bin/false

FG> You can add this file, make sure it is executable, and restart
FG> bootstrapping.

FG> I've filed https://bugs.gentoo.org/655290 for the automagic
FG> dependency.

In case anyone else is following along at home, the one amendment to
make to Fabian's workaround suggestion is that it turns out that just
having "fltk-config --api-version" return a failure error code is not
enough to make the configure test fail. The test is written as:

   FLTK_VERSION=`${FLTK_CONFIG} --api-version`
   if test ${FLTK_VERSION} != "1.3" ; then

so that failures of test are also not !=. So the fake fltk-config
script needs to print a number other than 1.3, not just return false.
But with that change, the workaround works as intended.

The other key debugging technique that I ended up figuring out on my
own is that you can test a bootstrap using a different version of the
Portage tree by pre-filling a $PREFIX/usr/portage directory (for
instance based on any commit from the repo/gentoo.git repository) and
touching a ".unpacked" file within it. Another useful piece of
information to know is that the Git commit corresponding to a Portage
tree is in $PREFIX/usr/portage/metadata/timestamp.commit .

In the time since, the bug that Fabian filed has also been fixed in a
new ebuild from the pinentry maintainer. I've checked that that fixes
things, by copying it into an older Portage tree (e.g., 6f597edb was
the one from my original experiment) and doing a successful bootstrap.

 -- Stephen