crash upon IVS (TL2016)

[dohyunkim]

With luaotfload v2.7-fix-2, compiling an example which contains IVS (Ideographic Variation Sequence) gives me an error.

\input luaotfload.sty
\font\jpfont=ipaexm.ttf at 10pt
\jpfont
\char"5307
\char"5307\char"E0100
\char"5307\char"E0101
\bye

The error message is

error: ...xmf-dist/tex/luatex/luaotfload/fontloader-2016-05-04.lua:19524: attempt to compare number with nil

[phi-gamma]

The error message is

error: ...xmf-dist/tex/luatex/luaotfload/fontloader-2016-05-04.lua:19524: attempt to compare number with nil

Thanks for the report. That’s inside code that the generic loader
does not (anymore) share with Context.

[dohyunkim]

For the record.
After my private e-mail to Hans, he changed the code in font-gbn.lua, in my view, too aggressively.
My intention was:

diff --git a/font-gbn.lua.orig b/font-gbn.lua
index daa072b..b6df93c 100644
--- a/font-gbn.lua.orig
+++ b/font-gbn.lua
@@ -126,11 +126,11 @@ function nodes.handlers.nodepass(head)
                             local variant = hash[getchar(p)]
                             if variant then
                                 setchar(p,variant)
-                                if not redundant then
-                                    redundant = { n }
-                                else
-                                    redundant[#redundant+1] = n
-                                end
+                            end
+                            if not redundant then
+                                redundant = { n }
+                            else
+                                redundant[#redundant+1] = n
                             end
                         end
                     end

whereas the changes made by Hans is:

diff --git a/font-gbn.lua.orig b/../Downloads/font-gbn.lua
index daa072b..a02406b 100644
--- a/font-gbn.lua.orig
+++ b/../Downloads/font-gbn.lua
@@ -126,17 +126,19 @@ function nodes.handlers.nodepass(head)
                             local variant = hash[getchar(p)]
                             if variant then
                                 setchar(p,variant)
-                                if not redundant then
-                                    redundant = { n }
-                                else
-                                    redundant[#redundant+1] = n
-                                end
                             end
                         end
                     end
+                    -- per generic user request we always remove selectors
+                    if not redundant then
+                        redundant = { n }
+                    else
+                        redundant[#redundant+1] = n
+                    end
                 end
             end
         end

This aggressiveness would be no problem, until we happen to encounter in the future an abnormal font which uses Variation Selectors in OpenType feature processing. As far as I know from my experience, however, there exists no such font. So I think it would not be bad to wait until someone reports an issue regarding the disappearance of Variation Selectors.

[phi-gamma]

···<date: 2016-05-07, Saturday>···<from: Dohyun Kim>···

After my private e-mail to Hans, he changed the code in
font-gbn.lua, in my view, too aggressively.

This aggressiveness would be no problem, until we happen to
encounter in the future an abnormal font which uses Variation
Selectors in OpenType feature processing. As far as I know from
my experience, however, there exists no such font. So I think it
would not be bad to wait until someone reports an issue regarding
the disappearance of Variation Selectors.

Agreed. Let’s just hope that font isn’t going to be a proprietary
one. As soon as we have a valid counterexample we can revisit the
matter.

Best,
Philipp

[aminophen]

As soon as we have a valid counterexample we can revisit the matter.

I'm also having trouble with IVS feature, and

• https://github.com/adobe-fonts/source-han-sans/tree/release/OTF/Japanese

we can use SourceHanSans-Regular.otf, which is freely available. My MWE:

\input luaotfload.sty
\font\jpfont=SourceHanSans-Regular.otf at 10pt
\jpfont
\char"5307
\char"5307\char"E0100
\char"5307\char"E0101
\bye

error: ...xmf-dist/tex/luatex/luaotfload/fontloader-2016-05-04.lua:19524: attempt to compare number with nil
.
\bye ->\par 
            \vfill \supereject \end 
l.7 \bye

[khaledhosny]

until we happen to encounter in the future an abnormal font which uses Variation Selectors in OpenType feature processing.

Many (may be most) Mongolian fonts handle variation selectors using GSUB substitutions instead of special cmap tables.

[dohyunkim]

Current issue does not affect Mongolian variation selectors (U+180x);
only standard variation selectors (U+FE0x) and Ideographic variation selectors (U+E01xx) are affected.