Re: Possible regression with `groff -C`

[G. Branden Robinson]

At 2020-04-12T11:12:32+1000, John Gardner wrote:
> After building Groff from master, I noticed one of Roff.js's regression
> tests failed when asserting compatibility mode:
> 
> *Command:*
> printf '.ds FOO BAR\n\\*[FOO]' | groff -C -Tutf8 | head -n1
> 
> 
> *Expected output:*
> FOO]
> 
> *Actual output:*
> BAR
> 
> 
> This only started happening since I installed a freshly-built Groff with
> the latest changes. Groff 1.22.4 (and earlier) don't have this problem.

I've never had a bisection go so smoothly.

Of course this breakage was my fault (a statistical likelihood).

d322f1951dcfc6ac7816fe8a654731041533443c is the first bad commit
commit d322f1951dcfc6ac7816fe8a654731041533443c
Author: G. Branden Robinson <address@hidden>
Date:   Fri Jun 28 01:37:44 2019 +1000

    tmac/unicode.tmac: Execute in compatibility mode.
    
    ...since the script uses a GNU extension (.char).
    
    Also update editor hints to Vim and (modern) Emacs.

:100644 100644 8e7973b62998f3db237618ed01ec9f3882fa83f9 
be9d68f2fde2d2f1a7551b81deb5ac0082f0e724 M      ChangeLog
:040000 040000 5b2239b52ef1d282f42800338a687eb6036a21b5 
31458639171a78e96a10d24cff5498956a7bab3b M      tmac

That description is ass-backwards.  What I did was add a line turning
compatibility mode OFF (.cp 0).  (The ChangeLog was already fixed some
time ago.)

But .char _is_ a groff extension, so, hmm.

diff --git a/tmac/unicode.tmac b/tmac/unicode.tmac
index 89864a6b..5bda095b 100644
--- a/tmac/unicode.tmac
+++ b/tmac/unicode.tmac
@@ -1,8 +1,14 @@
-.\" -*- nroff -*-
-.\"
 .\" unicode.tmac
 .\"
+.nr _C \n(.C
+.cp 0
 .char - \[hy]
 .char ` \[oq]
 .char ' \[cq]
-.\" EOF
+.cp \n[_C]
+.\"
+.\" Local Variables:
+.\" mode: nroff
+.\" fill-column: 72
+.\" End:
+.\" vim: set filetype=groff textwidth=72:

The following experiment on HEAD confirms the smoking gun:

$ printf '.ds FOO BAR\n\\*[FOO]' | ./build/test-groff -C -Tascii | head -n1
troff: <standard input>:2: warning: macro '[' not defined
FOO]

...since unicode.tmac doesn't get read if the output device is "ascii":

$ grep -C 1 utf8 tmac/tty.tmac
.
.if !'\*[.T]'utf8' \{\
.  ie c\[pc] \
--
.
.if '\*[.T]'utf8' \
.  mso unicode.tmac

Note that the entire contents of unicode.tmac appear in the diff above.
I do not even begin to have a theory as to what's going on here.

Any lurking experts have any suggestions?

Regards,
Branden

signature.asc
Description: PGP signature