Subject: regexp(3) improvement
Newsgroups: mod.sources
Approved: jpn@panda.UUCP

Mod.sources:  Volume 4, Issue 49
Submitted by: utzoo!henry (Henry Spencer)

One flaw of my regexp(3) as distributed is that there is no way to get
a literal `&' or '\n' (n a digit) past regsub().  The V8 manual page
made no mention of an escape convention.  It turns out that this is a
deficiency in the documentation rather than the software.  Accordingly,
the following update should be applied to my regexp(3) to preserve full
compatibility and to add this useful facility.

In regsub.c, change the following (line numbers approximate only):

67,69c67,71
< 		if (no < 0)	/* Ordinary character. */
< 			*dst++ = c;
< 		else if (prog->startp[no] != NULL && prog->endp[no] != NULL) {
---
> 		if (no < 0) {	/* Ordinary character. */
> 			if (c == '\\' && (*src == '\\' || *src == '&'))
> 				c = *src++;
> 			*dst++ = c;
> 		} else if (prog->startp[no] != NULL && prog->endp[no] != NULL) {

In regexp.3, add the following sentence to the end of the paragraph
describing regsub:

To get a literal `&' or `\e\fIn\fR' into \fIdest\fR, prefix it with `\e';
to get a literal `\e' preceding `&' or `\e\fIn\fR', prefix it with
another `\e'.

And add the following two lines to the "tests" file:

abcd	abcd	y	&-\&-\\&	abcd-&-\abcd
a(bc)d	abcd	y	\1-\\1-\\\1	bc-\1-\bc

My thanks to Mike Lutz at Rochester Institute of Technology for noticing
this issue and alerting me to it.

				Henry Spencer @ U of Toronto Zoology
				{allegra,ihnp4,decvax,pyramid}!utzoo!henry