Re: gnulib-tool bug

[Eric Blake]

Bruno Haible <bruno <at> clisp.org> writes:

> Ah! The culprit is the --posix option that gnulib-tool adds:
> 
> This looks like a bug to me: \\ expands to nothing. No way to produce a
> backslash in the replacement.

I concur.

> Can you confirm my interpretations?
> 
> > POSIX says that \/ is not portable when / is not the delimiter.
> 
> In addresses (selection of lines), certainly. But it the 's' command? I 
cannot find
> the wording that says so.

"The replacement string shall be scanned from beginning to end. An 
<ampersand> ’&’) appearing in the replacement shall be replaced by the string 
matching the BRE. The special meaning of ’&’ in this context can be suppressed 
by preceding it by a <backslash>. The characters "\n", where n is a digit, 
shall be replaced by the text matched by the corresponding back-reference 
expression. If the corresponding back-reference expression does not match, then 
the characters "\n" shall be replaced by the empty string. The special meaning 
of "\n" where n is a digit in this context, can be suppressed by preceding it 
by a <backslash>.  For each other <backslash> encountered, the following 
character shall lose its special meaning (if any). The meaning of a <backslash> 
immediately followed by any character other than ’&’, <backslash>, a digit, or 
the delimiter character used for this command, is unspecified."

Thus, 's,^,\/,', is non-portable.  But 's,^,\\/,' should result in adding the 
two bytes \ and / at the front of the string.  You are, however, correct 
that 'sed --posix' fails to do this.  And considering Solaris behavior, I think 
we have a definite GNU sed bug.  Paolo?

$ echo hi | sed --posix 's,^,\/,'
hi
$ echo hi | sed --posix 's,^,\\/,'
/hi
$ echo hi | /usr/xpg4/bin/sed 's,^,\/,'
/hi
$ echo hi | /usr/xpg4/bin/sed 's,^,\\/,'
\/hi

Time for me to document this in the autoconf manual.

-- 
Eric Blake