How groff should locate AFM files for gropdf (was: build system: devpdf/download regression)

[G. Branden Robinson]

Hi Deri,

At 2022-07-16T15:56:28+0100, Deri wrote:
> My understanding is that the selection of paths to search is appended
> to the value of the --with-urw-fonts-dir config variable (if any).

Yes.  To make it clear for mailing list readers, that's determined by
font/devpdf/Foundry.in, which currently looks like this.

$ grep ^foundry font/devpdf/Foundry.in
foundry||@urwfontsdir@:(gs):/usr/share/fonts/type1/gsfonts:/usr/share/fonts/default/Type1:/usr/share/fonts/default/Type1/adobestd35:/usr/share/fonts/type1/urw-base35:/opt/local/share/fonts/urw-fonts:/usr/local/share/fonts/ghostscript
foundry|U|@urwfontsdir@:/usr/share/fonts/type1/gsfonts:/usr/share/fonts/default/Type1:/usr/share/fonts/default/Type1/adobestd35:/usr/share/fonts/type1/urw-base35:/opt/local/share/fonts/urw-fonts:/usr/local/share/fonts/ghostscript:(gs)

> Then the sed in the configure script parses the string of paths as it
> does now.

This logic is in an Autoconf macro we define called
`GROFF_URW_FONTS_CHECK`.  I'll append its present definition from Git
HEAD to the end of this mail.

In detail, right now, _if_ both awk and ghostscript are available, the
search path scraped from the output of "gs -h".  Even if it is not
available, this search path is suffixed with a static list of
directories and finally with the ./configure'd --with-urw-fonts-dir
argument, if any.

This search path is used to look for a single representative URW font
which has gone by a variety of names (URW Gothic Book, a free clone of
Adobe Arial roman).  At present, we assume that this is a valid and
reliable indicator of the presence of all base 35 URW fonts.

The are currently a few problems: (1) this Autoconf macro (a.k.a.
configure check) is not run if either of the 'awk' or 'gs' commands is
not available.  This is a change I made, but it is too strict.  We only
use awk to parse the output of 'gs -h' if 'gs' is available, but we
would still have the static paths and, if used, the ./configure'd
--with-urw-fonts-dir argument, if any.

Problem (2) is that the search order used by the Autoconf macro is
inconsistent with Foundry.in above.  We can see that Foundry.in's
ordering is

(user-configured-urw-fonts-dir):(gs -h search path):(static list),

but GROFF_URW_FONTS_CHECK's is

(gs -h search path):(static list):(user-configured-urw-fonts-dir).

The order being inconsistent _might_ not matter, since the only thing
it's used for at configuration time is to decide the issue of whether
_any_ font metrics are avilable.  But I'm starting to have doubts,
because Arial roman (groff: 'AR') is _not_ one of the base 14 Type 1
fonts, an important factor you note below.  What do you think?

I suggest that a third problem, which Ingo raised if I recall correctly,
is (3) that we're hard-coding these "static paths" in two places,
violating the Don't Repeat Yourself principle.  We're already
inconsistent enough between these two files!  I propose instead to move
this list of hard-coded directory names to a standalone file in the
source tree.  This list can then be interpolated into a temporary shell
variable in GROFF_URW_FONTS_CHECK to conduct the search, and also
interpolated into the Foundry file generated during the build from
`Foundry.in`.  I'd replace the static list in the latter file with
the string "(static_list)" or similar, and have `BuildFoundries.pl`
perform the same substitution when generating `Foundry`.

> When Foundry file is built the @urwfontsdir@ is replaced by the string
> formed in the configure script. You can use ":" as the path separator
> since both sed and BuildFoundries currently use that character as a
> separator, it is only when parsing the ghostscript supplied paths that
> the O/S dependent @PATH_SEPARATOR@is used.

That makes sense.

> If this is done, all the static paths in the Foundry file can be
> removed leaving just the @urwfontsdir@ and (gs) entries.

Okay--I question this, though.  If we do that, then people without
Ghostscript installed will _have_ to ./configure --with-urw-fonts-dir to
have any chance of locating the URW fonts for gropdf operation.

But maybe that's okay?

> I don't think ghostscript releases the afm files any more, just the
> fonts, which is sufficient to provide the default foundry but cannot
> be used to generate the extended glyphs for the "U" foundry. If the
> ghostscript fonts are not found, or has the fonts in %rom%, (and no
> URW fonts either) you end up with build warnings and a degraded devpdf
> capable of using the base 14 fonts but not embedding anything. These
> are the three possible outcomes when BuildFoundries is run with an
> empty download file:-
> 
> ┌──────────┬─────────────────────────────┬───────────────────┬───────────┐
> │          │ Ghostscript missing (%rom%) │ Ghostscript found │ URW found │
> ├──────────┼─────────────────────────────┼───────────────────┼───────────┤
> │Base 14   │              ✓              │                   │           │
> ├──────────┼─────────────────────────────┼───────────────────┼───────────┤
> │Base 35   │                             │         ✓         │     ✓     │
> ├──────────┼─────────────────────────────┼───────────────────┼───────────┤
> │U Foundry │                             │                   │     ✓     │
> └──────────┴─────────────────────────────┴───────────────────┴───────────┘

Understood.

> So, BuildFoundries never fails,

The wrinkle here is that I have changed it so it _can_ fail, if it is
run with the `--strict` option, which I also do when it is run as part
of the groff build.  Right now, the only circumstance under which I make
the script fail is when the AFM file for a devps font description can't
be found.  Though I had planned to make _all_ warnings fatal!  Having
second thoughts.

It now sounds like that change was incorrect, because it prevents
support for the scenario in the upper-left hand corner of your table.
As you say, even there are no AFM files present, gropdf can still
produce valid PDF output if the document limits itself to the base 14
fonts and the user doesn't request embedding of all fonts (the `-e`
flag).  This will be common enough, for instance to render any groff man
page except those for grops(1) and gropdf(1), and probably countless
other man pages that don't perform stunt font selections.

Further, the automated test I just added

https://git.savannah.gnu.org/cgit/groff.git/commit/?id=ca0c6eb84bdcf08b518f4c70318f84fb91b863c8

requires all 35 PostScript fonts to have their description files copied
over to gropdf's device/font description directory (plus 'EURO', a font
we supply).  But if I understand you, in the no-Ghostscript or
Ghostscript-only-reports-%rom% scenario, this test will spuriously fail.
gropdf will be limited but it would still work for the base 14 fonts as
discussed above.

While I have your attention, I'd like to confirm for my own mental
health that it's _not_ necessary to copy over the SS (slanted symbol)
and ZDR (Zapf Dingbats Reversed) grops font descriptions.  groff has not
been doing this, and I assume that it's not necessary--that PDF has
primitives to mirror glyphs about the vertical axis and slant them if
necessary.  Is that correct?

> it generates whatever it can. If the configure script can detect the
> ghostscript fonts separately from the urw fonts then the appropriate
> message can be output.

Okay.  I need to think about what we should communicate to the user in
this case.  It sounds like we might need (urp) yet another `_NOTICE`
Autoconf macro, this one warning the user of the above "only base 14
fonts, no embedding" scenario.

What should happen if Ghostscript is useless (is not installed or
returns only "%rom%") but the user supplies --with-urw-fonts-dir?

Thank you for your help in thinking through these issues.

Regards,
Branden

# Check availability of URW fonts in the search path given by 'gs -h'
# supplemented with the paths where font/devpdf/Foundry.in expects them,
# or in the custom directory passed to 'configure'.

AC_DEFUN([GROFF_URW_FONTS_CHECK], [
  AC_REQUIRE([GROFF_AWK_PATH])
  AC_REQUIRE([GROFF_GHOSTSCRIPT_PATH])
  groff_have_urw_fonts=no
  if test "$AWK" != missing && test "$GHOSTSCRIPT" != missing
  then
    AC_MSG_CHECKING([for URW fonts in Type 1/PFB format])
    _list_paths=`$GHOSTSCRIPT -h | $AWK 'BEGIN { found = 0 } /Search path:/ { 
found = 1 } /^[ ]*\// { print $'0' }'| tr : ' '`
    _list_paths="$_list_paths \
      /usr/share/fonts/type1/gsfonts/ \
      /usr/share/fonts/default/Type1/ \
      /usr/share/fonts/default/Type1/adobestd35/ \
      /usr/share/fonts/type1/urw-base35/ \
      /opt/local/share/fonts/urw-fonts/ \
      /usr/local/share/fonts/ghostscript/"
    if test -n "$urwfontsdir"
    then
      _list_paths="$ _list_paths $urwfontsdir"
    fi
    for k in $_list_paths
    do
      for _font_file in \
        URWGothic-Book.t1 \
        URWGothic-Book.pfb \
        URWGothicL-Book.pfb \
        a010013l.pfb
      do
        if test -f $k/$_font_file
        then
          AC_MSG_RESULT([found in $k])
          groff_have_urw_fonts=yes
          break 2
        fi
      done
    done
    if test $groff_have_urw_fonts = no
    then
      AC_MSG_RESULT([none found])
    fi
  fi
  AC_SUBST([groff_have_urw_fonts])
])

signature.asc
Description: PGP signature