Re: make check: grub-mkrescue fails missing unicode.pf2. Leftovers in /tmp

[Glenn Washburn]

On Wed, 15 Feb 2023 09:51:18 +0100
"Thomas Schmitt" <scdbackup@gmx.net> wrote:

> Hi,
> 
> Glenn Washburn wrote:
> > IIRC, xfonts-unifont is pulled in by the unifont package on Debian,
> > but it is the one with the files you need. No, its not documented,
> > but then not every single debian package name needed is documented
> > because the documentation is partially distribution agnostic. If it
> > were documented it would be in ./INSTALL, which does talk about
> > needing unifonts. If you'd like to modify that file to help other
> > Debian users, I'm supportive.
> 
> Unicode is mentioned two times in INSTALL.
> Once for building "GRUB's graphical terminal (gfxterm)", once as
> "Platform-agnostic tools and data".
> 
> There is also a section "Debian named packages required mostly for the
> full suite of filesystem testing" which could host the name of the
> needed package. Although it is not about filesystems the mentioning
> could be justified by the word "mostly" in the headline.
> Or it could be mentioned as own point one level higher under
> "Prerequisites for make-check".

Yes, this makes sense.

> But actually the lack of unifont did not hamper building GRUB from
> git. So i would expect that this tolerance is somewhat reflected by
> the tests.
> 
> How about putting in tests/grub_func_test.in a test for the existence
> of /boot/grub/fonts/unicode.pf2 with an informative error message if
> it is missing ?
> Since we know that xorriso under grub-mkrescue will cause early
> failure if the file is missing, there is no use in running grub-shell
> with the demand for unicode.pf2.

Yes, I like this. And in your case you got a test failure, when you
should get a test error, which indicates that the test could not
actually be run and so was not a true failure.

> I see that grub_func_test.in already has provisions to declare
> "Functional test failure". Maybe the new code path could join that.
> Or it could be declared as skipped. (I still have to find out how to
> cause a SKIP message in the test log.)

A SKIP is caused by exiting with code 77. However, that's not what
should be done in this case.

I'll send a patch for this soon based on these suggestions.

Glenn