Re: "troff -a" is now communicating less information

"G. Branden Robinson" <[email protected]> Tue, 16 Dec 2025 19:31:31 -0600

Newsgroups	gmane.comp.printing.groff.general
Message-ID	<20251217013131.fuy2dmcktnf2eine@illithid>

At 2025-12-15T16:07:26-0600, Dave Kemper wrote:
> I noticed recently that, compared to previous groff releases up to and
> including 1.23, the latest groff built from git has less informative
> -a output in some instances.  Consider this two-line input file.
> 
> .fschar S \[trademark] \N'228' Enjoy a glass of
> Fluerma\fS\[trademark]\fP tonight.
> 
> Running this through "groff -a" has produced the below output from (at
> least) groff 1.19.2 through groff 1.23:
> 
> <beginning of page> Enjoy a glass of Fluerma<S trademark> tonight.
> 
> This output has now changed.  The fact that it HAS changed, I'm OK
> with: -a is billed as approximate output and subject to change.  But
> the nature of its change makes it, I think, less useful.  The -a
> output under the latest groff code is:
> 
> <beginning of page> Enjoy a glass of Fluerma<---> tonight.
> 
> This is straightforwardly enough less useful that I opened what I
> thought would be an uncontroversial bug report about it (
> http://savannah.gnu.org/bugs/?67817).  In particular, if a document
> has 100 such character definitions, they ALL get transformed into
> "<--->" rather than each into a unique string as they used to.
> 
> Branden has argued in favor of the change, in a way I won't attempt to
> summarize for fear of misrepresenting his position; you can read his
> response at that URL, or wait for his inevitable reply here.

I found the change responsible for the commit.

I was able to revert it cleanly (i.e., no merge conflicts).

You might be interested in what it was.

$ git show 3d249f79fd662e0e033ee1dc564f242b1a753e3c|cat
commit 3d249f79fd662e0e033ee1dc564f242b1a753e3c
Author: G. Branden Robinson <[email protected]>
Date:   Mon Mar 17 01:19:54 2025 -0500

    [troff]: Fix Savannah #55799 (`-a` opt + `char`).

    * src/roff/troff/node.cpp (composite_node::ascii_print): When rendering
      a composite node for approximate output, traverse into it and write
      its nodes' representations in forward order.

    Fixes <https://savannah.gnu.org/bugs/?55799>.  Thanks to Dave Kemper for
    the report.  Problem appears to date back to groff's birth.

Regards,
Branden

signature.asc (application/pgp-signature, 833 B)