+++ This bug was initially created as a clone of Bug #1813187 +++

Description of problem:
During investigating https://bugzilla.redhat.com/show_bug.cgi?id=1722900 was found the font mentioned in summary fails with mismatch on FcPatternGetInteger() from fontconfig when gs asks for FC_WEIGHT value.
I'm not sure whether the font or fontconfig is at fault, so I'm filing the issue on font package.

I'm willing to help with debug, if needed.

Version-Release number of selected component (if applicable):
abattis-cantarell-fonts-0:0.201-2.fc32.noarch
fontconfig-2.13.92-6.fc32.x86_64
ghostscript-9.50-1.fc32

Steps to Reproduce:
1. gs -dNODISPLAY -dBATCH nasmdoc.ps


Actual results:
You can see in logs a message:

DEBUG: FC_WEIGHT didn't match in /usr/share/fonts/cantarell/Cantarell-VF.otf

Expected results:
No such message

Additional info:

--- Additional comment from Kalev Lember on 2020-03-13 08:32:41 UTC ---

Sorry, I don't know much about fonts and don't think I can be of much help debugging this. Can you ask Jakub and/or Tagoh if they know what's going on (CCd)?

--- Additional comment from Nicolas Mailhot on 2020-03-13 08:42:52 UTC ---

On the fontconfig side, that’s basically bug
https://gitlab.freedesktop.org/fontconfig/fontconfig/issues/222

No idea if gs also contains problematic weight assumptions, invalidated by the introduction of variable fonts.

On the packaging side cantarell has probably not been converted to new font packaging guidelines yet. It needs some name fixing (use of a space-free ExtraBold qualifier, removal of Regular in Name ID 4 for the Regular face, making NameID 4 consistent with Name ID 17 for other faces). Though except for the ExtraBold part, everything else will going to be workarounded in future fontconfig versions.

--- Additional comment from Nicolas Mailhot on 2020-03-13 08:52:38 UTC ---

(Also fontconfig exposes a generic meta face for variable fonts, this face does not have any specific weight, it gives the weight scale for the whole family)

--- Additional comment from Akira TAGOH on 2020-03-13 08:58:05 UTC ---

That isn't related to this.

I don't know how ghostscript collects fonts from fontconfig though (probably using FcFontList()?), the problem here would be that they have dealt with a meta face as a normal face which has true in FC_VARIABLE and contains a weight in range but not integer.

That's not a bug in font nor fontconfig.

--- Additional comment from Zdenek Dohnal on 2020-03-13 09:25:03 UTC ---

Hi Akira,

yes, you're right, gs uses FcFontlist() - does the following to get font list:

-----------------------------------------------------
    /* load the font set that we'll iterate over */
    pat = FcPatternBuild(NULL,
            FC_OUTLINE, FcTypeBool, 1,
            FC_SCALABLE, FcTypeBool, 1,
            NULL);
    os = FcObjectSetBuild(FC_FILE, FC_OUTLINE,
            FC_FAMILY, FC_WEIGHT, FC_SLANT,
            NULL);
    state->font_list = FcFontList(0, pat, os);
-----------------------------------------------------

then when it enumerates the fonts it calls (f.e. for FC_WEIGHT):

result = FcPatternGetInteger (font, FC_WEIGHT, 0, &weight_fc);

(similar thing for FC_FILE, FC_FAMILY, FC_SLANT, FC_OUTLINE).

Would you mind telling me/pointing me to a manual how to deal with it correctly, if it is not a problem of font/fontconfig?

--- Additional comment from Akira TAGOH on 2020-03-13 10:07:42 UTC ---

If ghostscript has no plans to support variable fonts or a plan to support only named-instances, you can set false to FC_VARIABLE 
 as a query pattern; that should be more simple and easier than your patch applied in ghostscript; you can take care of the named-instances in VF font files as same as usual face then. otherwise set FcDontCare to FC_VARIABLE and need more changes in the code to deal with non-named-instances perhaps.

The meta face which you faced the issue this time provides such information instead of providing it as one of faces because there are so many patterns to enummerate everything and quite hard to do. the targetted properties such as FC_WEIGHT, FC_WIDTH and FC_SIZE has a value in FcRange which means they can use an unique value between them.

Unfortunately there are no more detailed document though, pango source code may helps how to deal with them.

--- Additional comment from Akira TAGOH on 2020-03-13 10:16:22 UTC ---

You can see what values the meta face provides. try fc-list -b -v Cantarell:variable=true for example.

--- Additional comment from Zdenek Dohnal on 2020-03-13 10:40:41 UTC ---

IMO they are not aware of the fact the variable fonts need a special care when are used from fontconfig.

I'll check with upstream about variable fonts then.

Thank you for the info!