Support request for pTeX-specific validation test

[aminophen]

I'm considering about introducing l3build test files into development repo of pTeX and friends. However, current l3build-check.lua strips some pTeX-specific log entries:

    -- Remove 'normal' direction information on boxes with (u)pTeX

    -- Remove '\displace 0.0' lines in (u)pTeX

I guess these are intended to reduce diffs between pTeX and Western TeX, but for us we need only pTeX-specific tests. Actually we'd like to know what would change in the process of pTeX refinement. (e.g. is '\displace 0.0' correctly inserted in the correct place?)

Is it possible to provide some way of not stripping off these logs?

[josephwright]

I'm not 100% sure I follow. My impression, purely from comparing pdfTeX and (u)pTeX logs, was that \displace 0.0 was always going to be a no-op, and that any 'real' information would be \displace <some-other-value>. Are there places that (u)pTeX doesn't add this at all, such that the difference is diagnostic?

It's of course possible to sort out selective normalisation: I can do that by checking if the default engine is (u)pTeX, and suppressing this normalisation if it is.

[josephwright]

Similar comment for yoko direction: all lines get a direction, so any that are not yoko will show?

[aminophen]

\displace 0.0 was always going to be a no-op

No, consider the following example for plain eptex:

\tracingonline1
\showboxdepth10000
\showboxbreadth10000
\setbox0=\vbox{\ybaselineshift 10pt\relax あaい}\showbox0
\box0
\bye

..\displace 0.0
..\JY1/mc/m/n/10 あ
..\displace 10.0
..\glue(\xkanjiskip) 2.40553 plus 1.0 minus 1.0
..\OT1/cmr/m/n/10 a
..\displace 0.0
..\glue(\xkanjiskip) 2.40553 plus 1.0 minus 1.0
..\JY1/mc/m/n/10 い

\displace 10.0 lowers "a" by 10.0pt, and \displace 0.0 returns the reference point (= goes up by 10pt) of "い".

yoko direction

\setbox0=\vbox{\tate あ\hbox{\yoko い}}\showbox0

..\dirbox(4.81108+4.81108)x9.16443, tate direction
...\hbox(7.77588+1.38855)x9.62216, yoko direction

I hope this kind of encapsulation can be tested as well...

[aminophen]

\displace 0.0 ... Are there places that (u)pTeX doesn't add this at all, such that the difference is diagnostic?

Previous versions (near TL2017 or older) did not emit \displace 0.0 in some necessary places, however, recent version of pTeX adds (or, should add) \displace 0.0 or \displace <some-other-dimen> everywhere.

yoko direction: all lines get a direction, so any that are not yoko will show?

For now I can't think of, but some old (near 2011?) version did not show yoko direcetion in some places due to a bug.

[josephwright]

What happens in TL'19? We are not keen on trying to cover multiple engine versions retrospectively ...

That said, the change here actually looks pretty trivial from the code point of view.

[aminophen]

pTeX and friends in TL19 still need some more refinements, which are raised as open issues on https://github.com/texjporg/tex-jp-build/issues. For now we are going to improve automatic insertion of glue penalties between Japanese and Latin characters (texjporg/tex-jp-build#85). That place is a candidate for insertion of displacement node, so removing \displace 0.0 from log means that we cannot use l3build for detecting what we actually changed.