[tex-k] Bug#814755: luatex font bug: "checksum mismatch" warnings on 64bit system

[Norbert Preining]

Hi Igor,

I reported this bug to Debian some days ago, but did not get a
response, so I'm writing here. Keep the CC please.

You expect quick response.
I am not sure, but it might be related to this discussion on
the fontinst mailing list:
http://www.tug.org/pipermail/fontinst/2016/001692.html

Norbert

------------------------------------------------------------------------
PREINING, Norbert http://www.preining.info
JAIST, Japan TeX Live & Debian Developer
GPG: 0x860CDC13 fp: F7D8 A928 26E3 16A1 9FA0 ACF0 6CAC A448 860C DC13
------------------------------------------------------------------------

[Akira Kakuto]

echo "(MAPFONT D 0 (FONTNAME cmb10))" > myfont.vpl
tftopl cmb10 >> myfont.vpl
vptovf myfont.vpl
rm myfont.vpl
vftovp -charcode-format=octal myfont.vf > myfont.ovp
rm myfont.tfm myfont.vf
ovp2ovf myfont.ovp
luatex '\font\f=myfont\end'

Here, I cannot reproduce the warning for both of 32bit and 64bit binaries
on Windows:

This is LuaTeX, Version beta-0.80.0 (TeX Live 2016/W32TeX/dev) (rev 5238)
restricted \write18 enabled.
No pages of output.
Transcript written on texput.log.

Also, "man ofm2opl" says about "-charcode-format" option,

Sorry, that describes on the legacy WEB version, now we call
it as 'wofm2opl'. Try wofm2opl --help.
In the C version, 'ofm2opl', the corresponding option
is -char-format.

Best,
Akira

[Karl Berry]

Also, "man ofm2opl" says about "-charcode-format" option,

Sorry, that describes on the legacy WEB version, now we call
it as 'wofm2opl'. Try wofm2opl --help.
In the C version, 'ofm2opl', the corresponding option
is -char-format.

Akira, if the only difference is the name of the option, I think we
should make the C version accept the -charcode-format too, as a synonym.
Why not? Can you do that? I'll adjust the man page to also talk about
them both. Thanks. -k

[Akira Kakuto]

Post by Karl Berry

Post by Akira Kakuto

Also, "man ofm2opl" says about "-charcode-format" option,

Sorry, that describes on the legacy WEB version, now we call
it as 'wofm2opl'. Try wofm2opl --help.
In the C version, 'ofm2opl', the corresponding option
is -char-format.

Akira, if the only difference is the name of the option, I think we
should make the C version accept the -charcode-format too, as a synonym.
Why not?

I was wrong. -charcode-format in 'wofm2opl' and -char-format in 'ofm2opl'
are similar, but they are not the same.

Best,
Akira

[Akira Kakuto]

FYI, C versions of ovp2ovf, ovf2ovp, opl2ofm, ofm2opl were probably
written by John Plaice or his students in order to replace corresponding
ones written by using the WEB. There are no man pages for them.
Sometimes bugs were found in them compared with the WEB ones.
Peter fixed a number of bugs. However the WEB versions still remain for the
safety by prepending 'w'. There may be few users today of Aleph/Omega,
however the utilities are still conveniently used, for example, to create
virtual fonts in (u)pTeX.

Thanks,
Akira

[Karl Berry]

Hi Igor,

checksum mismatch in local font cmb10 (-770990554 != -770990554)

This is complaining that a number is not equal to itself, clearly a
representation problem as you describe. I think you should report it,
with your analysis, on tracker.luatex.org. Perhaps it is already fixed
in the current code. The LuaTeX maintainers do not follow tex-k. -k

[Igor Liferenko]

Hi all,

Post by Karl Berry
This is complaining that a number is not equal to itself, clearly a
representation problem as you describe. I think you should report it,
with your analysis, on tracker.luatex.org. Perhaps it is already fixed
in the current code. The LuaTeX maintainers do not follow tex-k. -k

I reported it here: http://tracker.luatex.org/view.php?id=965

The response was that no problem was seen. But do get this warning. So
I investigated some more and discovered a strange thing: when luatex
is run from current directory, it loads only myfont.ofm, but when it
is run not from current directory, it tries to load also myfont.ovf
and cmb10.tfm (for the complete explanation see
http://tracker.luatex.org/view.php?id=965#c1580)

Why luatex loads different files depending from which directory it is
run? Is it something in TeX Live that causes this difference in
behavior, or it is the problem with luatex?

Regards,
Igor

[Karl Berry]

Hi Norbert,

I am not sure, but it might be related to this discussion on
the fontinst mailing list:
http://www.tug.org/pipermail/fontinst/2016/001692.html

No, Igor is reporting something unrelated, purely in LuaTeX's handling
of vf files. The fontinst issue that Michael and Lars were dealing with
was specifically about fontinst and arithmetic conversion in TeX of the
designsize value from the vpl (bad) vs. keeping it as a string (good).

I hope Lars releases a new fontinst with the fix. -k

[Karl Berry]

I reported it here: http://tracker.luatex.org/view.php?id=965

I'm glad Luigi could eventually reproduce the problem.

Why luatex loads different files depending from which directory it is
run?

I cannot explain anything about luatex behavior, which is (especially at
the moment) in a constant state of flux.

Is it something in TeX Live that causes this difference in behavior,

It is nothing in TeX Live, as far as I know. Nothing in TL in this
regard has changed, in any case.

or it is the problem with luatex?

I suppose so. All I can suggest is to report it to them. -k