Re: [PATCH 4/4] man/*.{1,1m,3x,5,7}: Soft-code ncurses version in `TH` call.

[G. Branden Robinson]

Hi Thomas,

I've root-caused the problem.

It was way easier than I thought.

At 2024-03-16T16:35:37-0500, G. Branden Robinson wrote:
> At 2024-03-16T17:17:03-0400, Thomas Dickey wrote:
> > troff: backtrace: '/usr/share/groff/1.23.0/tmac/an.tmac':512: while loop
> > troff: backtrace: '/usr/share/groff/1.23.0/tmac/an.tmac':516: macro 
> > 'an*scan-string-for-backslash'
[...]
> Yeah, that's my footer abbreviator all right (a new groff 1.23 feature,
> to say nothing of the suggestive path, so thanks to RMS for once for
> mandating the versioned subdirectory that so many people hate).

Here is the macro in question.

.\" Iterate through concatenation of arguments as a string.  If a bare
.\" backslash is found, make `an*string-contains-backslash` true.  Our
.\" caller should delete this register when done with it.
.de an*scan-string-for-backslash
.  nr an*string-contains-backslash 0
.  nr an*index 0
.  length an*max-index \\$*
.  while (\\n[an*index] < \\n[an*max-index]) \{\
.    ds an*char \\$*
.    substring an*char \\n[an*index] \\n[an*index]
.    ec @
.    \" Use a weird delimiter to reduce lexical colorizer confusion.
.    if _@*[an*char]_\\_ .nr an*string-contains-backslash 1
.    ec
.    if \\n[an*string-contains-backslash] .break
.    nr an*index +1
.  \}
.  rm an*char
.  rr an*max-index
.  rr an*index
..

Experienced *roff users will spot the problem instantly.

.    ec @

I'm changing the escape character to an implausible value so that I can
scan a string for a backslash, which is the _default_ *roff escape
character.

My idea of "implausible" was...an at sign.

Evidently my imagination was failing me badly when I wrote this, because
@-bracketed sed replacements are something we use _IN GROFF_.

Even in `TH` macro calls!

.TH groff_man_style @MAN7EXT@ "@MDATE@" "groff @VERSION@"

But I never stepped on my own land mine because I _never_ run the
formatter on man pages before all text transformations on them are
complete.

There is no such thing as index(3) or strchr(3) in *roff.[1]  To scan
the string for an '@' first and avoid it, perhaps using '#' instead, I'd
have to write a whole 'nother macro much like the above, or add a layer
of indirection that risks making things even more baffling.

So like Bartleby the Scrivener, I would prefer not to.

How badly do you need to keep running your "doit" sanity-checker on the
non-final versions of the man pages?

Do you feel that this is a real bug, or just noise?

Regards,
Branden

[1] I do actually have plans in this department.

    https://savannah.gnu.org/bugs/index.php?62264

    But I don't know if I'll have time to land them in groff 1.24, so
    other people will likely encounter this issue.  They have to turn on
    a non-default warning category and use the '@' sign in the 4th
    argument to their `TH` calls to hit it.

signature.asc
Description: PGP signature